CGreciano's Blog

AWS CLF and SAA notes and flashcards updated and refreshed

2026-07-10T00:00:00Z

A bit over a year ago I passed the AWS Cloud Practitioner (CLF-C02) and the AWS Solutions Architect Associate (SAA-C03) certifications. Upon doing so, I published my notes and flashcards of these certs. Now, one year later, I have refreshed my CLF-C02 materials, updating a lot of reference images and adding, editing, or deleting AWS service features and limits. Since the first 2 sections of my SAA notes are taken straight from my CLF-C02 notes, those sections have also been updated in SAA. Many things can change in one year in AWS/cloud technologies, and I’d like to delve a bit deeper on this topic in this blog post. But if you just wanted to get the updated materials, you can just stop reading and just go to the links:

Certification	Online Notes (Notion)	PDF Notes	Anki Flashcards
AWS CLF	CLF Notion Link	CLF PDF Link	CLF-C02 Flashcards
AWS SAA	SAA Notion Link	SAA PDF Link	SAA Anki Link

Note that if you purchased any of the paid materials before the updates, you can download the updated version from Ko-Fi completely for free!

The Biggest Problem with Cloud Study Materials: Updates

Cloud technologies evolve and change quite fast. Not quite as fast as modern AI technologies (which are changing at an insane pace), but still fast enough to make training materials age like milk. For example, while updating my materials for the CLF cert, the following features and facts have changed for Amazon S3 in the past year:

An S3 object used to have a maximum size limit of 5TB. This limit is now 50TB. Handy for those huuuuuge video files!
Although S3 buckets live in an AWS region, it was a well-known quirk that when creating an S3 bucket you had to give it a name that was globally unique. If it so happened that someone else in the world had named their bucket exactly like you wanted, you had to (annoyingly) find a new, not-yet-taken name. This is finally no longer the case! If you now use account regional namespaces, you can create an S3 bucket with a name that must only be unique to your account and region. In order to keep its ARN globally unique, AWS appends a suffix (which is unique to your account and region) to your bucket’s name under the hood. Wonderful!
S3 buckets live within a region… except in the case of the S3 One Zone-Infrequent Access storage class. As the name indicates, buckets of that class live only within a single AZ. Well, there’s a new storage class called S3 One Zone-Express. In the same vein, buckets of this storage class live only within a single AZ, and they’re called Directory Buckets. The appeal of this new storage class is that it offers really high performance for S3 operations, which is very handy for stuff like intensive data apps or ML training.

S3 objects being 5TB max and S3 buckets requiring globally unique names have been established facts for years, but they’re no longer true. Most AWS training materials will however not reflect these changes, unless they receive regular updates. This includes the study notes that a kind Redditor shared 4 years ago, the free flashcards uploaded to AnkiWeb that have the most upvotes but were uploaded 6 years ago, and yes, even video courses by popular instructors. If I had to plot the usefulness of cloud training materials versus time, it would look something like this:

So in essence, training materials might hold good value for 1-2 years, but then start to drastically lose value. If they are old, you will need to cross-reference them with AWS documentation to make sure you’re learning the latest facts and figures.

Always Check Date of Latest Update, or Learn to Fact Check

The lesson to learn is to make an effort to learn with updated materials, or get into the habit of fact checking old materials. For example, Andrew Brown has plenty of AWS certification courses on YouTube in the FreeCodeCamp channel. It’s easy to determine how old the courses are because YouTube registers the date of when a video is uploaded. And Andrew Brown himself performs periodic refreshes of his AWS courses because they do indeed get outdated.

I have to shout out Stéphane Maarek and his video courses in Udemy. He does an outstanding job of keeping his courses up-to-date with the latest AWS trends and news, better than any other instructor I have seen. In fact, one way you can quickly check what has changed in a certain AWS certification is to redownload his course slides after a while. His slides have version numbers and you’ll see how after a while the version will be higher. Then, you can upload both the old slides and the new slides to https://www.ilovepdf.com/compare-pdf and compare both documents. A semantic comparison will reveal what is new and what is no longer relevant!

Updating my Materials

I currently offer study notes and flashcards for 5 AWS certifications. While it’s true that I created them when studying for each of those certs, I don’t want to be yet another person who shares and never updates. I hope to dedicate some time on a yearly basis to keep them updated, but it will of course be a best effort considering this is a side-hustle and not my full-time job. I’m also more self-conscious now that if I add more cert materials in the future, that will add to my workload in the future to keep them updated. 😅

I will always be transparent on what version number my materials are, and each version update will include a log of what the updates are (just like software apps!). If you ever purchase any of my materials in my Ko-Fi shop, you are entitled to get the updated versions for free once I release them (you must log in to Ko-Fi with the same user that purchased my materials, and redownload the purchased materials to get the updated version).

I met my first work colleague in Finland F2F for the first time

2026-06-18T00:00:00Z

April 2021. I had just moved to Finland after getting a consultant job at Netlight. My first day at my new job was also the first day working for my client Fortum. I was hired as a Java backend engineer. New job, new client, new home, new country… and remember that this was in the middle of the Covid pandemic! 😅

Fortunately, there was another Netlight consultant in the same client and project, who had been there for quite a while already. Juho provided invaluable support as I got to gripes with the new environment, easing my mind on the stuff I didn’t know or fully understand, while encouraging me on the stuff that I did get right. And looking back, 5 years ago my hard skills were quite lacking: I was decent at programming in Java and Python, but my Spring Boot and Git skills were mediocre, and I had no clue about RabbitMQ, AWS, or even Docker, all of which were heavily used in that project. Juho left Netlight just 2 months after I joined, and we never had the chance of meeting in person due to the pandemic, but even that short time together was important for me to adapt to a project where I stayed for almost 2 years.

Fast-forward to today. A bit over a month ago I was looking for a new job, and one of my interview processes with a certain company progressed quite well. I did my due diligence with the LinkedIn filters and it turned out that Juho had worked at that same company some years back! I reached out to him and asked about his experience. He kindly answered back, I thanked him, and asked him if he’d be up for a lunch sometime to catch up. We did indeed meet up: today we had lunch at a nice Thai restaurant. This turned out to be the very first time that I personally met my “first colleague in Finland”… 5 years after we worked together! 😂

Whenever I feel like networking is intimidating, I remind myself that reaching out doesn’t need to be awkward or complicated. Be genuine, and life will open up opportunities naturally. Some of these events will turn into funny anecdotes, like mine. 😁 Also, there’s not enough gratitude in our world. You can never thank enough. I made sure I told Juho how much I appreciated his support, even after all these years. Everyone (including you) will feel better if you’re grateful, I promise!

I just joined Nitor as Senior AI Engineer

2026-06-10T00:00:00Z

I just started working as a Senior AI Engineer in another software consulting company: Nitor. Back in January of this very same year I started my stint at Wave Company, and I was not expecting it to be only a few months long, but here we are. New beginnings are always exciting and interesting, but they also normally require adapting to drastic changes. I’m grateful that employers continue to trust in my skills and services, even in a difficult market for software engineers.

Consulting in Finland

I have spent most of my professional career in software consultancies. Only between 2018 and 2020, when I worked for Samsung, did I work for a product house. For the past 5 years I have lived and worked in Finland, in four different consulting companies: Netlight, Reaktor, Wave Company, and now Nitor. There are still plenty of others here in Helsinki, should I “catch 'em all” like a Pokémon trainer? 😂 Honestly I hope I stay at Nitor for a while, I’d like some stability at this point in my life.

Consulting in Finland is quite different than, for example, consulting in the US. The horror stories I hear and read about consultants working overtime in order to get that last PowerPoint slide ready by 23:00 are not what happens in Finland, a country where (thankfully!) work-life balance is respected much more. We’re more like loan engineers over here, where we get assigned to projects and teams in clients and work as (mostly) regular employees there.

I sometimes wonder why Finnish companies hire consultants to do the work that a regular software engineer would do, since consultants are more expensive. A big part of it is probably that Finnish law makes it really difficult to fire employees after their probation period. With consultants, you can hire them temporally and let them go at any moment. This means that we consultants might live on the edge and with more stress than our in-house engineer counterparts, but in exchange we usually get a better salary and more benefits from our employers. I have also noticed that it’s easier for me to end up in consulting companies because their hiring process is smoother than regular companies (who take forever to hire people in a very cumbersome process, often lowballing you if they do make an offer).

First Impressions at Nitor

My colleagues Antti and Amir together with me on our first day at Nitor, company of G.O.A.T.s

I have barely been 10 days in Nitor, but there is already one nice thing and one less nice thing I can say about working here.

👍 Nitor Iron Bank

I feel that work equipment influences employees directly, for better or for worse. When a company provides high quality devices and machines for their employees to do their job, they feel valued and cared for. And if instead the company gives you the cheapest gear that barely allows you to do your job, you feel strictly like a pair of hands. It is of course a delicate balance between giving an employee everything they need and want to be happy without overspending or provoking jealousy in other employees who didn’t get the same specs.

This is why I have been impressed with Nitor’s solution: the Nitor Iron Bank. Apart from evoking fond Game of Thrones memories of the Iron Bank of Braavos, this solution to employee equipment is really elegant. Employees start with a generous budget, and they are free to choose pretty much any work equipment and gear that they want within that budget. The budget then accumulates slowly every month, so by the time you need to repair or renew your devices, you can use your budget again. If at any point you’re out of budget, you can access used and loan devices on emergencies. Simple, yet elegant: employees are happy to choose what they want, but they also need to be responsible with their allocated budget.

👎 Nitor uses MS365 Workspace

The vast majority of software companies use one of two enterprise workspaces: either Google Workspace or MS365 Workspace. In my opinion Google’s Workspace is vastly superior to Microsoft’s. And yet MS365 is used a ton in the industry. Sadly, Nitor uses the MS365 Workspace. I’d rather not use Microsoft products if I can help it, but looks like I don’t have much of a choice here. A lot of Nitor consultants share my tastes: in their latest developer survey Microsoft products were in general not very liked, with Microsoft Teams being the most hated product of all. 😂 And as it was to be expected, getting access to my Microsoft Entra ID was NOT a smooth experience! In my current setup, I need to run two different browsers: Mozilla Firefox in order to authenticate with the Microsoft Entra ID, and Brave Browser because it’s my current browser of choice (but it has problems to authenticate with Microsoft Entra ID, check out this GitHub issue). Yay for hybrid workflows! 😂

Nitor Rebranding

Coincidentally, Nitor has decided to rebrand itself exactly at the same time as I have joined the company. We’ve gone from blue and purple shades to dynamic red and black shades. The logo has been simplified, but the firey torch is still present. “Nitor” means “brightness, brilliance, striving” in Latin, and the torch comes from a Greek mythology story of how Prometheus stole the fire from the gods. The idea is that Nitor is here to illuminate the world in the software industry via cutting-edge technology and excellent delivery. That’s a brand I certainly feel identified with! 😀

My study notes and flashcards for AWS GenAI Developer Professional AIP-C01

2026-05-30T00:00:00Z

LINKS:

Notion notes (FREE): https://psychedelic-cuticle-e74.notion.site/AWS-Generative-AI-Developer-Professional-AIP-C01-2b586c7395e780b2b18ce55c72874333
Notes in PDF: https://ko-fi.com/s/9a30d68cde
Anki flashcards: https://ko-fi.com/s/22e0104816

I just published my study notes and flashcards for the AWS Generative AI Developer Professional (AIP-C01) certification! 🎉 If you have been following me in my journey to get this certification, you’ll remember that I obtained this certification back in January, and that I wrote a detailed blog post about it. Back then I promised to polish my notes and flashcards. I have been very busy at work and it has taken me a bit longer than expected, but I have finally done so! 😄 If you purchased the early beta version of my flashcards at a discounted price, you can now download free of charge the updated and complete version in my Ko-Fi shop.

About my Notes for AIP-C01

The AWS AIP-C01 exam validates a candidate’s ability to effectively integrate foundation models (FMs) into applications and business workflows. This certification demonstrates practical knowledge of how to implement GenAI solutions into production environments by using AWS technologies.

My notes cover the knowledge domains comprehensively. You will dive deep into Retrieval-Augmented Generation (RAG), vector stores, and Agentic AI, and learn the nuances of AWS products in the context of GenAI. Amazon Bedrock is the main focus of the exam, particularly Bedrock Knowledge Bases, Bedrock AgentCore, and Bedrock Guardrails. There is a free version of my notes which is hosted online in a Notion site, and there is a paid PDF version which you can purchase in my Ko-Fi shop (useful for printing, studying offline, etc).

My materials assume you have already passed the preceding AWS AI certifications, namely AWS AI Practitioner (AIF-C01) and AWS ML Engineer Associate (MLA-C01). If you haven’t done so yet, I really recommend you do so, because the professional-level AIP-C01 builds upon those. You can find notes and flashcards of those preceding certs also in my website if you need them!

The main source of my materials is the Udemy course by Frank Kane and Stéphane Maarek for this certification, although I have also included some information from other websites and AWS blogs. I took this video course when I prepared for the exam, and I can highly recommend it to complement my materials. Both Kane and Maarek are highly regarded instructors in the AWS community. Note: The video course contains a lot of content from preceding certifications which I have not included in my materials, since I already have that content in my other notes.

About my Flashcards for AIP-C01

As many followers know, I like to complement study notes with flashcards. You need to include active ways of learning in your studies and flashcards are a pet favorite of mine. I have created Anki flashcards for each section of my study notes, and they all include tags and references so that they are easy to find and filter. You can purchase them in my Ko-Fi shop.

No AI-slop Guarantee!

Although this is an AI-related certification, I crafted my study notes and flashcards manually. These materials are NOT AI-generated! There’s a plethora of AI-generated study materials out there, but these are not one of them. I have crafted and reviewed them myself.

Wrap-Up

This is not an easy exam (check the aforementioned blog post for my detailed experience). Make sure you also take practice exams and get used to the lengthy and draining questions and answers of AWS professional-level certifications. That said, this is a satisfying exam to pass. And as a Senior AI Engineer, I can also say that the domains tested in this exam (RAG, vector stores, agentic AI…) are very relevant in our current industry!

If you use my materials and find them helpful, be sure to tag me online when you post in social media that you passed the exam, and I’ll be happy to congratulate you! There’s a particular joy in collecting testimonials from happy peers! 😁

The Wild West of corporate AI roles

2026-05-25T00:00:00Z

AI continues to change and shape the software industry to this day. Companies are slowly starting to adapt to the new AI landscape in 2026, and many AI-related jobs are starting to pop up. Good news: AI has created new jobs. Bad news: these are mostly roles for seniors (hopefully junior roles will soon make a comeback). Additional bad news: many companies don’t yet know how to correctly define these roles, and many still struggle to understand how AI can bring value to them (if it even can!) and how much it actually costs to build AI stuff.

The AI landscape still feels like the Wild West in many aspects, so let me attempt to briefly describe the most common AI roles we’re seeing on demand right now. This can be useful to know how to position yourself, or to know what kind of person you actually need in your team. This is my POV as someone who has been involved in AI projects in the past 2 years, and who also has several years of experience in the software industry and in consulting.

AI Strategist

Tons of companies and business owners want to introduce AI in their products and business, but they don’t know how. And even if they believe they know what they want, they usually don’t know how much it actually costs to get it. If you have never had an AI specialist in your team or company, or if you have already tried to embed AI into something and been disappointed, please hire an AI strategist.

This is a business-facing role: expect lots of meetings, conversations with stakeholders, (pre-)sales conversations, chasing and keeping up-to-date with AI trends, managing AI-hype and expectations, educating employees and stakeholders about AI, setting budgets, suggesting skillsets and roles to build AI products… This is already a full-time role, and AI strategists are NOT expected to build anything! They simply don’t have time for it! Hence, they don’t necessarily need to come from a technical background. They shouldn’t be tech-illiterate however: without a strong conceptual understanding of data architecture and cloud systems they risk falling for vendor snake oil. AI strategists are experienced in working with teams of AI engineers, defining requirements for AI products, communicating constantly with stakeholders, and managing expectations. This is a very similar role to that of a Product Owner in the well-known Scrum framework.

AI Engineer

AI engineers can also be described as AI developers or AI software engineers. FWIW this is my current role. They are software engineers specialized in building and maintaining AI systems. They know what LLMs are from a theoretical PoV, their strengths and limitations, and they know how to develop and build systems that leverage existing AI models (colloquially, you can call these systems “ChatGPT wrappers”). They know AI paradigms and architectures like Retrieval-Augmented Generation (RAG), Model Context Protocol (MCP), AI agents… They know how to build AI tools and apps, scale them for enterprise and production use, secure them, and maintain them. They bring their full-stack and backend development skills and fit all the Lego pieces together to make sure your AI system runs smoothly.

Don’t expect an AI engineer to be an AI strategist or an ML engineer (see below). Some people can do any of these roles, but if you expect them to be an AI strategist and an AI engineer at the same time, they’re not going to be able to deliver very well. Choose one or the other role at a given time. Or hire one of each. Just as you would hire a software engineer to build stuff and not to sit in meetings the whole day, don’t hire an AI engineer to sit the whole day in meetings concerning AI. They won’t have time to build what you asked.

Difference between ML and AI engineers

Before the irruption of LLMs and Generative AI three years ago, the concept of “AI engineer” was drastically different than nowadays. AI engineers back then were the ones who built AI models from scratch, trained them with data, fine-tuned them, implemented ML algorithms… They did what is currently known as “traditional AI”. Nowadays these are known as “ML engineers”, and the distinction should be clear. Many job postings are still listing “ML/AI Engineer”, or “AI Engineer” with knowledge requirements for traditional AI. Sometimes they ask for a PhD in ML/AI. Make sure you clarify whether the role is for an ML or an AI engineer, because not everyone can do both. People with PhDs in ML/AI often lean more toward research and experimentation rather than enterprise software engineering, meaning they might need a strong software engineer by their side to scale their work. AI engineers can build the scalable software surrounding AI, but they might struggle to train and fine-tune AI models.

Image source: this LinkedIn post by Louis-François Bouchard

AI Lead

Instead of hiring a lonely AI engineer and burden them with tons of responsibility, it’s usually a good idea to hire multiple AI engineers and create an AI team. One of those engineers, maybe the most senior, could then be the AI lead. They have the last word on architectural decisions, and they usually spend more time reviewing code and coordinating team efforts than actually coding or building. This is a common pattern in software development teams (one engineer is usually the dev lead).

Other AI Roles

LLMOps engineer, MLOps engineer, Whatever-AI-Ops engineer: These are to AI engineering what DevOps engineers are to software engineers. They are in charge of AI operations: CI/CD pipelines, infrastructure to support the AI systems, setting up monitoring and observability frameworks… These are technical roles that can blend with AI engineers in the same way that DevOps blends with software engineering.
AI designer: UI/UX designers specialized in AI products. AI products are software products and they also need intuitive and user-friendly UIs.
AI consultant: If you’re in consulting, you can very well be an AI consulting. You might need to wear several different hats for different AI roles we have covered here, depending on what your client needs. For your sanity, insist in wearing only one hat at a time, or bad things will happen. If you’ve been a software consultant, you’ll understand how you might need to be a frontend developer, DevOps engineer, or a Scrum Master depending on your client needs. Just don’t be all at once!

Wrap-Up

Role	Primary Focus	Core Skillset	Technical Depth
AI Strategist	ROI, business value, expectation management	Product management, AI trends, budgeting	Low to Medium (Conceptual)
AI Engineer	App integration, pipelines, scaling models	Full-stack/Backend, RAG, MCP, API usage	High (Software Dev)
ML Engineer	Model creation, fine-tuning, training	Mathematics, data science, deep learning	High (Data/Research)

This is by no means set in stone. These roles will surely continue evolving, and AI will continue to create jobs (hopefully some entry-level ones for those junior graduates!). It’s also interesting to see how the big AI vendors adapt their offering in 2026, since it’s well-known that they’re running at big losses and won’t be able to be generous any more in the near future. I suspect that those engineers who can be both effective and efficient in their AI usage will be in high demand soon. Cost-efficiency and token-optimization are the next big survival skills for AI engineers.

Did I miss an AI-related role that should have been in this article? Any description disagreements? Reach out to me in whatever social media platform I’m in and let’s talk!

How to study and filter Anki decks by topics and tags

2026-05-15T00:00:00Z

Many people who have downloaded or purchased my Anki decks have asked whether it’s possible to adjust the order of their flashcards to their liking, or to study by topic, theme, or AWS service. The short answer is YES! You certainly can do all of this by using Anki tags and filters. In this blog post I will teach you how to do so via the Anki desktop app, using my AWS AI Practitioner (AIF-C01) deck as an example.

Filtering by Anki tags or fields in the Anki browser

In a previous blog post I explained how to import an .apkg file into your Anki app. Let’s assume you’ve already imported this deck and that you can see it in your list of decks. Click on the deck, and then click “Browse”. This will open the Anki browser, showing all the cards of your selected deck.

Observe how each flashcard has at least one tag associated to it, as well as a field called Reference, with one or more links to the sources I used to create the flashcard. If you are creating your own Anki flashcards, I strongly suggest you add at least one tag and one reference to each of your cards, this brings a lot of benefits in the long run!

You can filter the list of cards shown in the browser by providing a tag or text in the text field at the top. Append label:Amazon-Comprehend after "deck:AWS AI..." and hit Enter. You will then only see the flashcards labeled with Amazon-Comprehend. You can achieve the same result by selecting the appropriate tag in the Sidebar filter on the left side of the browser:

You can also filter by text: append any text to the search bar and Anki will show you only the cards that contain that text in any of the card’s fields. This is obviously very useful to find, for example, all cards that mention one concept or entity. But it can also be useful to, in our case, filter cards by source or lecture. Filtering by links, however, is a little tricky: adding a full link (e.g. https://www.udemy.com/course/aws-ai-practitioner-certified/learn/lecture/44886687 ) to the browser will fail to find the cards, because Anki’s search function uses non-alphanumeric characters as modifiers, and those are present in URLs (e.g. :, /, etc). So, while it’s useful to keep the full URL in Reference to quickly find the source of the flashcard, to filter all flashcards with the same URL you must select only a relevant sub-sequence of alphanumeric characters inside the URL. In our case, select the lecture’s ID (the sequence of numbers) and voilà! You can see all flashcards that belong to the same lecture:

Studying cards by tag, topic, or lecture

By default, when you start studying an existing deck, Anki will show you the flashcards in the order that the author originally inserted them in the deck. If you use one of my decks, the cards will be shown following the order of topics in my study notes. So in the case of my AWS AI Practitioner (AIF-C01) notes, here is the table of contents:

If you follow the default order, you will first study AI 101, then Cloud Computing Fundamentals, then AWS Accounts, then Bedrock, etc. But what if you decide to start by studying Amazon Q? In that case you’d rather skip the flashcards on Bedrock and Prompt Engineering and start answering cards about Amazon Q. In this case I recommend that you suspend ALL flashcards in the deck, then filter the cards you want to study, and unsuspend only those cards. Anki only shows you unsuspended cards when it’s time to study, so you can progressively unsuspend the flashcards of the material as you cover it. Step-by-step:

Load your deck in the Anki browser
Select all cards in the deck: Top Menu -> “Edit” -> “Select All”, or use the keyboard shortcut (Cmd + A on Mac or Ctrl + A on Windows/Linux)
Right-click on the cards -> “Cards” -> “Toggle suspend”
You have now suspended all cards in the deck. If you type is:suspended in the Anki browser search (or click on Suspended card state on the filters to the left), you should see all cards of the deck, since they have all been suspended. If you now tried to study the deck, Anki would tell you there is nothing to study, because there are no new unsuspended cards available!
Just as we learned in the previous section, filter the deck with tag:Amazon-Q (or selecting the tag in the filters of the left panel). You should now only see the cards tagged with Amazon-Q. Select them all -> Right-click on cards -> “Cards” -> “Toggle suspend”
You have now unsuspended all cards tagged with Amazon-Q in the deck. Since these are the only cards that are unsuspended, Anki will now only quiz you on the Amazon-Q cards. Neat!

You can follow a similar process to later unsuspend cards from a different tag, from a specific lecture, or from any other filter you can concoct.

Anki Filtered Decks and cramming

So far we have learned how to study a deck in an order that suits us via tags and filters. This is with a long-term learn & review mindset, which Anki and Spaced Repetition algorithms (SRS) excel at. But what if you want to cram, let’s say, all the SageMaker flashcards the night before the AWS AI Practitioner exam, as a final review? Well, I was about to skip this use case, because last-minute cramming has tons of drawbacks (increased stress, makes your pre-exam sleep worse, what you haven’t learned by then you probably haven’t learned anyway…). But if you insist on ignoring my advice (which is the advice of cognitive science BTW! 🤓), Anki does offer a feature to create temporary filtered decks. These are subdecks containing cards from a tag or specific filter, and those cards are returned to their parent/home deck once you review them in your filtered deck.

How to create a filtered deck?

Select your deck, then click “Custom Study” on the bottom.
Select “Study by card state or tag”. You can choose the number of cards and whether you want some or all cards.

Click “Choose Tags”.
Check the checkbox “Require one or more of these tags”. Then select the tag(s) you want to study (e.g. Amazon-SageMaker)
Click on “OK”. Anki will have now created a filtered deck with all SageMaker cards, and you’ll see this as a “Custom Study Session” deck (click “Study Now” to start).

There are tons of things you can do with filtered decks but frankly I myself haven’t used them much (because I avoid last-minute cramming, ahem! 😏). I suggest you consult the Filtered Decks section in the official Anki manual if you want to delve deeper. There are even add-ons like AnkiCram if that’s your sort of thing (but I don’t endorse them myself).

What if I use Anki on mobile?

Indeed, this tutorial has focused only on the desktop app of Anki. This is intentional: the desktop app offers much more functionality than the mobile apps, and you should use the desktop app whenever you’re tweaking or configuring anything in Anki. I’m not 100% sure if everything I have shown here can be done on mobile, and it might even be so that it’s not the same in iOS or Android. So I’m gonna keep this tutorial simple and just for desktop. I myself only use the Anki mobile app for reviewing my cards, and I suggest you do the same: desktop Anki to create, edit, and tweak cards and configurations; and mobile Anki to review cards.

Conclusion

You now have the skills and knowledge to filter Anki cards by tags, topics, and lectures, which means you can study Anki cards in a different order than the default one. Although it can feel a bit overwhelming at first, Anki is a super powerful tool, and learning some basic tweaks can grant you a lot of flexibility and personalization. This tutorial should also have convinced you to always tag your Anki cards and also include a reference URL to your source(s).

Have Anki tags and filters been useful to you in a different way? I’d like to know! Reach out to me or comment on my YouTube video, I’d be curious to hear!

My very first email address is gone forever

2026-04-30T00:00:00Z

My very first email address was richan5@eresmas.com. The password to access it was simply “ajedrez”, which means “chess” in Spanish. I don’t mind writing this publicly because I just found out that my first email account no longer exists. I hadn’t used it since I created my current Gmail account back in 2008, almost 20 years ago. And yet it felt as if one of my first footprints on the internet is now gone forever. In fact, I’m not sure I could find an older footprint.

My first venture into the internet

I can’t remember if I was 11 or 12 years old. I think I was 11 years old. I was already into video games, computers, and strategy games. I had been so since I was old enough to sit in front of a screen and input stuff via mouse and keyboard. But the internet was still a fascinating new concept to me. I had played StarCraft (my favorite video game ever) already for a few years, and with the internet I would be able to play with people in a remote location. That thought honestly felt magical. But back in 2000 internet at home was far from mainstream in Spain. The computer room in my school had just received internet access, and it was heavily restricted. A few of my schoolmates already had internet at home, but I wasn’t so lucky. One of those schoolmates who did was the one who showed me how to get my first email address.

He suggested to grab a free email address from EresMas, a popular ISP in Spain back then. I chose “richan” because my best childhood friend had nicknamed me “Richan” years ago (I realized how cringey that was when I discovered that “-chan” in Japanese is a suffix denoting a small child or a female character 😅). And “ajedrez” as my password since I already loved chess back then. This email address allowed me to reach out to people who created and edited StarCraft fan-made maps, and thus I made my first online friends. Remembering all of this is quite nostalgic honestly. Soon I created a second email account, richan10@eresmas.com. My reasoning was that I wanted different addresses for different things (not sure if I didn’t know about filters back then, or if email filtering was not yet a thing, can’t really remember!). I created these addresses back in school, before I ever got an internet connection at home (which I finally got when I was 14 years old).

A screenshot from EresMas back in 2001. Source: this tweet

Eresmas, Orange, and Gmail

EresMas eventually got acquired by Orange, another Spanish ISP, and access to those email addresses was routed to their web portal. I created yet a third email address back then, richan@orangemail.es. And while I used these three addresses for years, refusing to ever create a Hotmail or Yahoo account (I barely touched MSN Messenger!), at some point I did try out Google’s new email client, Gmail, and was blown away at how good it was. The UI was so clean and intuitive! And there were no embedded ads and much less spam.

That’s right, spam became a big problem in my original email addresses. As a teenager I rarely cared about privacy. Although my schoolmate had educated me on how I should obscure my personal data and never give my real data to strangers (I thank him for that!), giving my addresses out freely meant I got a looooot of spam. And while Gmail had good spam filters, Orange did not. So yeah, almost 20 years ago my original addresses became this disposable bin and an email I could give out more freely (e.g. when requiring a one-time login to a website), while I was more careful with my Gmail accounts. It was a primitive way of having the benefits of email aliases back then.

Evolution of Online Communication

Email was the primary way of communicating online back when I took my baby steps into the internet. A few years later internet forums became all the rage and I spent so many evenings in them. To be honest many teens also used MSN Messenger, but I preferred the async nature of communication. Some years later Facebook and social media became the norm. And nowadays that social media sucks, most people have moved on to Discord servers. Email still exists, of course, but it’s used mostly for transactions, newsletters, and formal communication.

These days I’m still using Gmail even though Google has turned much worse as a company in the last decade. With an ad-blocker, I’m free of publicity, and Gmail lets you configure the UI to your liking as well as tons of settings. I have given Proton mail a try, but despite the benefit of privacy, it doesn’t feel like a comfortable email client, and I can’t quite follow a GTD approach with it. Gmail has also spoiled me into not wanting to pay for email services, but I reckon I might need to change that perception at some point in the future.

Orange closes down their email services

Orange exists to this day, but to my surprise, they closed the EresMas and OrangeMail email domains some months ago. If you can read Spanish, here is the official announcement, and here is an excellent article describing how many Spaniards had email addresses from their ISPs back in the early 2000s. Names like EresMas, Wanadoo, Mixmail are now historic names and a legacy of the old internet in Spain. While my old email accounts were mostly spam honeypots and had been so for decades, I’m a bit sad that I will never be able to log in to them ever again, and maybe rescue my very first interactions with online folks. I know there’s nothing of major importance there, but it feels like losing some dusty and bad quality photos from your childhood, if you know what I mean. A lesson to be learned: nothing in the internet will be forever, and you should choose an established email provider like Gmail, Hotmail, etc over an email service from an ISP, which might get merged or bought out later down the line.

What were your early internet adventures like? Have you had similar experiences to mine? Reach out if you did!

Loved the Music of The Lion King, but not it's AI-generated scenes

2026-04-16T00:00:00Z

This past weekend I attended a concert titled “The Music of The Lion King” in Helsinki. I had been looking forward to the event for weeks, since The Lion King holds a very special place in my heart. Ever since my childhood, I have loved every bit of the movie, and it remains to this day one of my favorite movies of all time. A big part of that is its majestic music, with artists such as Elton John and Hans Zimmer. You might then understand why I was so excited to come to this concert. It’s not The Lion King Musical, but I was still hyped.

The concert delivered when it came to the performers and the music. The singers were amazing, their voice really powerful. The musicians and orchestra brought tears to my eyes on several occasions. The familiar music touched my heart. The concert also included a homage to Africa and some interesting facts surrounding The Lion King’s impact worldwide. Heck, we even got Shakira’s “Waka Waka” shoehorned in, and we were invited to join and enjoy the show. 😀 I was delighted to get a picture with the performers after the concert.

But there was a stain. During the performance, we had several screens showing images and animations related to the songs and The Lion King. And let me emphasize the “related” word. They were not the original scenes from the movie. They probably didn’t have the rights from Disney to show those in the event. Instead, we got clearly AI-generated videos of live-action animals resembling those of The Lion King. Sometimes they were OK to watch, other times the color palette and the soulless AI-slop were extremely distracting. During the interpretation of the Stampede theme, the wildebeests appeared to be carelessly treading instead of running for their lives. The hyenas seemed to have an ocean inside their mouths as they drooled excessively. And the animal bodies often had disproportionate limbs and features. The coloring in the scenes was also overbloated.

All of it naturally led to the key question: “would it have been better to NOT have any content at all over having AI-generated content?”. It’s a tough question to answer, and it will of course vary from person to person. In my humble opinion, I would have preferred no visual nor animated content at all than this AI-generated slop. And I know that some people worked on putting that animation together, I know that I’m complaining about someone else’s work. But the fact is that sometimes I had to close my eyes to more fully enjoy the concert and not get distracted. And what bothers me is that this drop in quality is just going to permeate in every form of media moving forward. I don’t mind quality content made by AI. But I do mind AI-slop being portrayed as quality content.

Have you been in any events recently where AI really turned you off? What is the best way, do you think, of showing dissatisfaction towards this trend (assuming we don’t want to go to the extreme of avoiding quality events just because they throw AI into them)?

Run your Local AI in less than 15 minutes! (no tech skills required)

2026-03-30T00:00:00Z

It’s 2026 and hence you’re most likely already using AI tools such as ChatGPT, Claude, or Gemini. You’re probably also worried about the rising costs of using these AI tools and there’s good reason to be: vendors of these tools are rising prices and/or introducing ads to try to be profitable. If you are entrenched on AI workflows and becoming dependent on a certain vendor, what will you do if the vendor spikes their prices to something you can no longer afford?

The antidote for your worries is using local AI! By running a free and open-source AI model in your machine and hardware, you can say goodbye to monthly subscriptions and ads! You might be scared of trying to install local AI on your computer and think it’s a complicated setup, but I assure you, it is surprisingly simple to do so, as I will show you in this blog post. If you’re here just for the instructions, jump directly to the Step-by-step installation of your local AI section. Before that section I want to first explain the tradeoffs between using local AI and cloud AI and then briefly present the AI stack of tools we will use to deploy AI models locally.

Tradeoffs between local AI and cloud/SaaS AI

SaaS stands for “Software-as-a-Service”: many of the online apps you use, like Gmail, Netflix, Spotify… are examples of SaaS. ChatGPT, Claude, etc are also SaaS, AI SaaS to be specific. Let’s be very clear on the pros and cons of running your local AI versus using an AI SaaS, i.e. a vendor’s AI hosted in the cloud. If you run your local AI, you are hosting the AI model(s) in your own hardware, which means:

👍 No subscription fees and no ads. Compare using Netflix or having video files locally on your computer: if you have the video file of your favorite movie in your machine, you can reproduce it as many times as you want from now until forever, without having to consume ads or paying a subscription to Netflix.
👍 No internet dependency → low latency (super fast transmission), can work offline. The AI is in your hardware. Your requests to it and its responses to you travel almost instantly within your machine, whereas accessing an AI SaaS means traversing the internet. This advantage is really noticeable if your internet connection is unstable and/or slow.
- ‼ Note that this advantage does NOT guarantee that your local AI will always be faster than an AI SaaS! If your local AI needs to “warm up”, if it’s bottlenecked by resources in your machine, or if you’re using a slow reasoning model, that extra waiting time might be higher than waiting for responses and requests to traverse the internet.
👍 Much more data privacy. This is very important for many users. While you can disable cloud AI from using your data to train their models, your data is still traversing the internet and is still being stored in the vendor’s servers. What’s worse, cloud AI models can learn enough about you and store sensitive stuff about you in their long-term memory. What guarantee do you have that vendors won’t sell your data to third parties for targeted ads (or worse things)? Notice how social media knows you scarily well? By using a local AI, your data stays in your machine, and thus stays private.

At the same time:

👎 Your hardware puts limits on how powerful your local AI can be. Some AI models can be very demanding and resource intensive. Some require multiple GPUs if you want to self-host them, which are expensive (you, like most people, might only have your laptop and nothing else). Think how a cinema most likely has better sound, lighting, and atmosphere for watching movies than your local living room with your run-of-the-mill TV (but you gotta pay cinema tickets if you want to use it).
👎 Related to the above: AI SaaS are recommended if you use AI a lot in mobile devices. Smartphones and tablets generally have much less performant hardware than computers and laptops, so running local AIs on mobile devices is tricky (you usually have to settle for SLMs, i.e. Small Language Models, as opposed to LLMs or Large Language Models). You can of course set up a local server in your home with a self-hosted AI and have your mobile devices at home access that local server. But that gets into home-lab territory and I promised to keep things simple for non-techies in this blog post.
👎 Some AI vendors have excellent AI tools that go beyond simple chat interfaces. Stuff like NotebookLM, Claude Code, Claude Cowork… setting alternatives to these products locally is not trivial. If you use them a lot, it makes sense to pay for them, but if you mostly just chat with AI models, then you can most likely get away with local AI!

Local AI Stack we will install

We will only install a few programs to run AI locally. If you’re non-technical, you might not have heard of these yet, so let me briefly introduce you to them:

Ollama: this is a backend service that runs quietly in the background. With Ollama you can download open-source AI models easily, and once downloaded they are exposed and available for use.
Open WebUI: this is a free and open-source webapp that acts as a frontend to the AI models exposed via Ollama. It provides the chat interface that loads in your internet browser, which is really similar to the web pages of ChatGPT, Claude, etc.
Docker: many techies like me love Docker containers. Think of them like bounded and sealed boxes where you can isolate and automate certain programs and apps so that they don’t interfere with other programs and apps in your machine. It’s like putting ice cream into a small container so that eating it won’t get messy (as opposed to ice cream in cones, which CAN get messy!). We will install a Docker container that comes bundled and integrated with both OpenWebUI and Ollama, greatly simplifying the installation process.

Image source: Mega Cone Creamery

This proposed stack is really simple, but there are many ways and alternatives to run AI locally. If you’re interested, I will suggest some alternatives later in this blog post.

Step-by-step installation of your local AI

Download Docker Desktop (the version that works in your OS)
Open Docker Desktop, then open a Terminal within Docker Desktop (it should be on the bottom right corner of Docker Desktop, click on “Terminal”, and grant permissions to open one)

Open this link in your browser (it’s Open WebUI’s official GitHub, focused on the section where you can install Open WebUI bundled with Ollama via a single docker command).
Copy the command to run a Docker container with Open WebUI bundled with Ollama. You most likely should use “CPU only” (but feel free to experiment with “GPU support” if you know what you’re doing). If using “CPU only”, the command is: docker run -d -p 3000:8080 -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:ollama
Paste that command in the Terminal of your Docker Desktop and hit enter. This will install Ollama and Open WebUI via Docker, which might take a while (several GBs of download).
Once it’s ready, you should see the open-webui container running in Docker Desktop (a green dot indicates it’s running). Notice that you can always stop, restart, or delete this container (which contains your local AI) via Docker Desktop (stop ⏹ and play ⏯ buttons to the right)

Open WebUI is ready to be loaded in your browser! Click on the “Port(s)” link in Docker Desktop for the container (should say 3000:8080), or navigate to http://localhost:3000
Create an admin login for Open WebUI (don’t worry, this login is needed only for local use, and it stays local).

Once created, login as an admin into your local Open WebUI, you will be greeted with a pop-up, which you can close. You will then see a chat interface similar to ChatGPT and other cloud AIs you might have used.
Before chatting with an AI model, we must first download it! And now comes the million-dollar question: which model should you download and use locally? Well, even though I will suggest a model in a moment, the annoying answer is “it depends”. This is like choosing which T-shirt to buy: often you need to go to the store and try on several different T-shirts before you find the one that fits you, feels comfortable, and you like how it looks. You can explore tons of different open-source models in Ollama’s website and see which one best fits within your machine.

Photo by Ashita Mata on Unsplash

All of that said, I suggest you start with Meta’s Llama 3.2. This is a smaller model that does a good enough job at the reasoning tasks that many people use ChatGPT for (summarizing text, drafting emails…), and is likely to run well in most modern laptops. If you have older or less powerful hardware, you might need to look for an even smaller model (you will know this if Llama 3.2 runs sluggishly or not at all). If Llama 3.2 runs super smoothly, you can explore more powerful models (Llama 3.2 is far from cutting-edge at this point, I just wanted to suggest a simple model to start off with)
To install Llama 3.2 in your computer, click on your Admin account name within Open WebUI (bottom-left corner in Open WebUI) > then go to “Admin Panel” > go to “Settings” > go to “Models” > click on “Manage” > enter llama3.2 in the field for pulling models from Ollama > click the “Pull Model” button > wait for the model to download (will take a few minutes)

Once the model is downloaded, you should see it under “Models” (you might not see it immediately, but you can wait, or refresh the page, or leave and come back to “Models”)

Now that our model is downloaded, we can chat with it! Go to “New Chat” on the top-left corner, and talk with your local AI!

Alternative stacks/tools for local AI

While the Ollama + Open WebUI stack we have used for Local AI is really straight forward to install and use, there are alternatives out there:

LM Studio: Best for those who want a visual “App Store” for local AI models. It’s a polished app that looks like a professional chat interface.
Jan: Open-source alternative to LM Studio which focuses on being a 100% offline personal assistant. If you’re a privacy-conscious user who wants a clean, simple UI, give it a try.

Wrapping up

I hope I convinced you to give local AI a try. Running local AI is much easier than what most people think it is, and it will definitely be worth it once cloud AI becomes too expensive and annoying. There is no need to vendor-lock yourself to AI tools!

I’d love to hear from you if my blog post inspired you to try out local AI and you liked it! And even if you disliked it, you can still reach out and tell me why! Good luck in your endeavors!

I passed AWS Machine Learning — Specialty (MLS-C01)

2026-03-17T00:00:00Z

As I explained in a previous blog post, when AWS announced their new Generative AI Developer Professional certification (AIP-C01), they also informed that they would be discontinuing the Machine Learning Specialty certification (MLS-C01). It’s going away soon (31 March 2026 is the last day one can take it), so I made an effort and decided to get it before it’s no longer possible. Last week I took the exam and passed! 🥳

Reasons to take this discontinued certification

Why get this certification if it’s going away? I had several reasons to do so:

The cert is still valid for 3 whole years. I won’t be able to renew it once it expires, but for the next 3 years the cert is valid and is a testament that I know about ML in AWS.
It was my first ever AWS Specialty certification exam and I was curious to probe its difficulty.
I had already passed the AWS Machine Learning Engineer Associate (MLA-C01) and the AWS Generative AI Developer Professional (AIP-C01) certs before. This ML Specialty certification had a massive overlap with the content from MLA-C01 (about ~85% of the content), so I barely had to learn new things to pass ML Specialty, just review a bunch of material.
ML Specialty asks you generic data science and ML questions that are vendor agnostic and not only AWS-related questions, so in a way I also got generic data and ML knowledge validated.

Impressions and Results

This ML Specialty exam felt slightly harder than MLA-C01, and much easier than AIP-C01 (to be fair, I did take AIP-C01 during its beta period, which makes it extra challenging). The results also seem to corroborate my feelings, I achieved quite a nice score while I barely passed AIP-C01 two months ago:

All in all I’m happy I got this cert, it was quite a low-hanging fruit for me since I already had done other AI certs for AWS. Unlike for the other AWS certs that I hold, I won’t be releasing study notes or flashcards for ML Specialty, since soon nobody will be able to take the exam anyway. But if you’re curious and want to tackle the challenge within the next two weeks, refer to my study notes and flashcards for MLA-C01, as I said 85% of the material is the same (just the focus of the exams is slightly different).

ML Specialty focuses exclusively on traditional ML and deep learning (DL) and does not validate your knowledge of modern Generative AI. I understand why it gets discontinued since GenAI is all the rage right now. Traditional ML and DL appear to be so much more niche these days in the industry. Still, I’m saddened by this trend. Generative AI got where it got thanks to traditional ML and DL. There are still plenty of problems in the industry that are much better suited for ML/DL than for GenAI, yet so many companies and projects are trying to shoehorn an LLM where it’s not optimal. It was also quite unique to have an AWS cert that asked you plenty of vendor agnostic questions, the new AI certs pretty much quiz you only on their AI products… Oh well. Hopefully we get to see this kind of cert in the future from AWS again!