A podcast version of my personal website. All the episodes are generated from a text to speech model.
Reading the feed…
I share why Pi—a minimal, model-agnostic coding agent—has become the backbone of my daily workflow across terminal, Tmux, and cloud-hosted agents. I introduce pi-tools, a collection of lightweight extensions I built to keep multiple Pi sessions manageable, including pi-jumper for fast session switching and pi-helicopter for a lean infobar, plus a few small utilities that stay true to Pi’s minimal, predictable philosophy. Relevant links: Original article Pi Mario Zechner Claude Code OpenAI Codex OpenCode Claude Code source leak (InfoQ) Session portability pi-llama (llama.cpp) How I self-host in
Source composition and averages. I'm proud to announce Bagaço v3, the third version of the largest European Portuguese pretraining dataset for large language models. Why a new version? After my work on Ginjinha, one thing became pretty apparent: Bagaço was large and diverse, but lacked high-quality data. I believe Bagaço …
July was busy — like always — but we made sure to take some time and touch some grass. Or mountains in our case. With the high temperatures across Europe we went to the Trentino region and hiked around the Dolomites for five days. Cool temperatures and beautiful scenery. Vitto's idea of …
For the past year, I've been very curious about the intersection of Large Language Models (LLMs) and the European Portuguese language. I've built evaluations, classifiers, and eventually the largest pretraining dataset in European Portuguese. Now, that's all fine and dandy, but it raises the obvious question: Can we pretrain an …
This month's Retrospectiva comes a little later than usual. The whole house got sick battling the flu the past couple weeks. Mixed with a small heatwave here in Copenhagen, means summer hasn't blessed us like we expected. At least not yet. Now that I'm feeling a bit better, I finally …
In January this year, OpenClaw went viral. You probably heard about it. My guess is some of you even use it - or at least tried to. I certainly did. I remember it clearly. I installed it, and very quickly noticed I was now spending time debugging a cron job created …
If there's one thing Retrospectiva does for me - other than giving me an excuse to write for you of course - it's making me realise how fast time moves. It's June already - and May was – well – hectic to say the least. Data Makers Fest was awesome - and I absolutely love Porto …
About 6 months ago I wrote a small rant on how open source TTS models still sucked. 6 months later, I'm happy to report that isn't the case anymore. January this year, Qwen, the famous Chinese AI lab, released Qwen3-TTS, an open-weights series of TTS models. The release included 2 …
Educational score over time vs. document count for Bagaco v2 A couple of months ago I released Bagaço - a pretraining dataset for European Portuguese. The idea was simple: take the FineWeb 2 dataset, limit it to web pages that look like they came from Portugal, and classify them into categories …
My Hermes agent – Saramago – giving me coaching feedback and looking for the ideal flights for an upcoming trip. Another month bites the dust. We've been back in Denmark for the past couple of weeks. I often joke that these are the only months in the year where Copenhagen is actually …
Update July 2026: The AMÁLIA models have now been released! You can find the weights/datasets on Hugging Face. Some parts of this blog post might be out of date. So beware! In December 2024, the Portuguese government announced AMÁLIA: a 5.5 Million Euro investment on a large-scale LLM …
Well, now would you look at that. Article 100 on this website. Ten years of writing on this small corner of the web. I'm writing this month's newsletter from my favourite place: the airplane. Happy Easter if you celebrate. With clients spread across time zones, I didn't really get the …
As you get older, life becomes complicated. Not in a bad way. There’s just more going on. We don’t all want to run marathons. Maybe you want to run a parkrun. Maybe you want to gradually increase your volume. Maybe you don't want to run at all. Whatever …
Given some text in Portuguese, how easy is it to determine if it's from Brazil or Portugal1? For native speakers, this is pretty easy – it's almost a feeling. But for machines: not so much. This might seem like a useless problem at first. But in the age of language …
Happy February! Just like that, we are close to wrapping up the first quarter of 2026. After quite some time roaming around, we finally flew back home to Copenhagen. It's cold, windy, and grey, but it's also calm, organized, and cozy. Most of all: it's home. Another good thing about …
Let's say your goal is to train a Large Language Model only on European Portuguese. Where do you start? What datasets are out there? What websites are being scraped for the large black box? Bagaço - named after the popular Portuguese moonshine - is a small step in that direction. In June …
Well—that was quick. Just like that, the first month of the year is gone. Winter blues would normally peak around this time. But not this year. We're spending time with family in my favorite place on Earth, somewhere along the Adriatic coast of Italy, in the Marche region. It's …
Let's face it. It might just be the year of agents. If you work in tech and your workflow hasn't changed in the last year or so - you're probably doing something wrong. For those of you who have. It's fun. We're building more than ever before! But there's a dark …
I get up feeling sleepy. I lace up my running shoes and head out the door. I fire up my favourite podcast. "This show is brought to you by..." I hate ads. I understand the attention economy. Companies are capitalizing more and more on everyone's time. I have nothing against …
And just like that, it's the end of the year again. We did not expect to make it to the South for Christmas this year. Getting a passport for a newborn is a painful, bureaucracy-filled process, especially for a Portuguese-Italian baby born in Denmark. But somehow—magically—the Italian embassy …
Benchmarking calorie prediction for Taralli There's something very funny about the current Machine Learning and AI landscape. If you're in the field you probably heard about it. "Vibes" they call it. When someone wants to test something out, they conduct a "vibe test". I call bullshit. How are you supposed …
Big news in the state of Denmark. And no, nothing's rotten. Allegra just came into the world. As she takes a nap, I take the opportunity to write November's Retrospectiva update. The most relevant thing this month is probably the release of my recent book about DeepSeek. It's nice to …
Back in May this year, my longtime friend Alex reached out and asked me if I wanted to collaborate with him on a book about DeepSeek. I would love to tell you the story of how I thought long and hard before getting back to him. I didn't. I just …
or at least the open versions of it. I have this very stupid rule. A couple of years ago I decided to turn this blog into a podcast. At the time, I decided to make up a stupid rule: whatever model I use to clone my voice and generate article …
Welcome to Retrospectiva. Retrospectiva is a monthly update about what I’ve been up to. In the age of LLMs, I’ve heard many argue that it’s hard (and useless) to write anything at all anymore. When anyone can prompt a model and get some text from the magic …
