AI Engineer project
Build a document Q&A tool grounded in retrieval
A RAG app that answers only from the uploaded document and cites its source — proven by asking it something the document doesn't cover.
The brief
Build an app that lets a user upload a PDF or text document and ask questions about it. Chunk the document, embed the chunks, retrieve the most relevant ones for a given question, and generate an answer that cites which chunk(s) it came from. The app should visibly decline to answer, rather than guess, when the document doesn't contain the answer.
Suggested stack
What you hand in
- A public GitHub repository with the running app and a README
- The README documents the chunking strategy and embedding/retrieval approach
- A short section showing an example where the app correctly declines to answer an out-of-document question
Grading happens against the rubric below, so read it before you start — not after.
How this is graded
Published in advance and weighted out of 100. Nothing here is a surprise.
For a range of test questions, the chunks actually retrieved are genuinely relevant to the question, not just the top-k by default similarity with no sanity check.
Answers cite the specific source chunk and don't state information the document doesn't contain — proven with an out-of-document test question.
Chunks are sized and split sensibly (e.g. respecting paragraph or section boundaries) rather than a fixed character count that cuts sentences and tables mid-thought.
A user can upload a document, ask a question, and get an answer with a working citation link back to the source, without needing to read the code.
No console errors or crashes, no broken layout, no leftover placeholder text or commented-out code.
Why this project is worth your weekend
- Retrieval-augmented generation is the single most in-demand AI-engineering pattern right now, appearing in nearly every real AI product.
- "Does it hallucinate" is the first thing any interviewer will test live, by asking a question the document doesn't answer.
- Citations back to the source chunk are what turns a toy demo into something a real team could actually ship and trust.
Where people lose points
- Sending the top-k retrieved chunks to the model with no relevance check, so a bad match still gets confidently answered.
- Chunking at a fixed character count that splits sentences and tables mid-thought, degrading retrieval quality.
- Answering questions the document doesn't cover as if it does, instead of visibly declining.
Other AI Engineer projects
Two or three of these turn an empty resume into a portfolio.
Built it? Get it scored against this rubric.
Submit your work and get a score on every criterion above, written feedback, and three resume bullets you can use straight away.