GroundWork

Document-grounded Q&A with verified citations and source-level evidence.

Inspiration

We just wanted to solve a simple problem — AI gives answers, but sometimes you have no idea where that answer actually came from. So we thought, why not make it show the source too?

What it does

GroundWork lets you upload documents and ask questions about them. It finds the relevant parts, uses Gemini to generate the answer, and shows citations to the actual text. If the answer isn't in the document, it can just say that instead of making things up.

How we built it

We used Java, Spring Boot and React. For searching the documents, we built our own BM25 retrieval instead of using a vector database. Gemini handles the answer generation and verification, while everything stays behind our backend API.

Challenges we ran into

Honestly, getting the Gemini integration working took more time than expected. We had API key issues, model availability problems and a few errors that kept showing up at the worst possible time. Making sure the answers were actually supported by the documents was another challenge.

Accomplishments that we're proud of

We finally got the whole thing working:

Upload → Retrieve → Gemini → Verify → Answer → Citation

Seeing an answer come back with the exact document chunk supporting it was probably the most satisfying part.

What we learned

We learned that making an AI answer something is the easy part. Making it answer from the right information and show why we should trust it is much harder.

We also learned way more about APIs, debugging and working under a deadline than we expected.

What's next for GroundWork

We'd like to add PDF and DOCX support, improve the retrieval, and make the source/citation experience even better. Basically, we'd like to keep working towards making AI answers a little less of a "trust me bro" situation.