Five projects that prove you can engineer AI, not just wire an API
One more chatbot proves nothing: wiring a model to a text box, anyone does it in an afternoon. This kit takes the five systems of a list video ([Instagram reel DfZiMLtSaca](https://www.instagram.com/reel/DfZiMLtSaca/)) and turns them into five steps, each with what it teaches, the catalogue brick and the test that says whether it worked: a document assistant that shows its source, a voice receptionist that books in a real calendar, invoices read and validated, a research assistant whose citations hold, and the evaluation platform that measures the other four.
Why
An AI portfolio is judged on what the system does when things go wrong, not on the demo that works. The video says it plainly: AI engineering is processing data, retrieving the right information, wiring tools, validating outputs, evaluating performance and surviving failures. Each of the five projects isolates one or two of those skills and makes them checkable. The fifth, the evaluation platform, is the one almost nobody builds: that is why it stands out.
When
You are looking for an AI engineering role or contract and your public repositories only hold chat assistants. You want a progression that fits in five repositories and tells well in an interview. Not for you if you are looking for a product idea: these projects exist to prove a level, not to find a market. Not for you either if you want 93 ideas sorted by difficulty, that is what the AI…
Tools included
- Rag Vector
- Pgvector
- Whisper
- Eleven Labs
- Unlimited Ocr
- Zod
- Rag Agentique
- Firecrawl Mcp
- Deepeval
- Ai Engineering Hub