About Workfreak

Why I built it

I was that guy. Watch a 2-hour Huberman video, feel pumped, close the tab. By morning it was gone. I had like 10 apps: goals here, food there, workouts somewhere else. None of them talked. I hated it. I wanted one thing where it all connects and the stuff I actually watch turns into stuff I actually do.

So I built Workfreak. Paste a link or drop a PDF. The AI pulls out the good bits, suggests goals, makes todos and habits from it. The Mentor sees my goals, my food, my workouts, my check-ins. It gives advice that fits my life, not the usual generic stuff. When I ask it something, it knows my context.

Now I watch something and I don't just forget it. I connect it. I get a plan. I actually do it. That's the whole point.

For the nerds

Technical stuff and what I learned building it

It's RAG. Your content gets turned into embeddings, stored, then pulled back when you ask. The Mentor doesn't guess. It retrieves. Your goals, your food log, your workouts, your knowledge sources, all flow into one context. The system prompt is long. You have to guess. That's how it sees everything at once.

What I learned: Chunk size matters. Too small, context fragments. Too big, retrieval gets noisy. Same with the prompt. You start short, then edge cases pile up. Cost control is real. The hard part wasn't the model. It was wiring the pipeline so the right bits surface at the right time. That took a while.

If you're building something similar: start with the data model. Get the relations right. The rest is RAG and polish.