All projects
ai agents
A retrieval-augmented question answerer
Answer questions over your own documents, with citations and a way to catch hallucination.
Why it matters
RAG is the most common way LLMs actually ship. The valuable skill is evaluating retrieval quality rather than assembling the pipeline, which is a weekend.
What you build
- Documents chunked and embedded sensibly
- Retrieval you have measured
- Answers that cite their sources
- An evaluation set that catches hallucination
If you want more
- Add hybrid keyword plus vector retrieval
- Add re-ranking
Build it
Chunking decisions dominate retrieval quality more than the model does.
- Generate answers that cite the chunks used
Material for the whole build
Shows you can build with LLMs while staying honest about what they get wrong.
Get a plan built around projects like this