OneSource

Inspiration

Many organizations struggle because their knowledge is scattered across tools like Google Drive, Slack, and GitHub, making it hard to find reliable answers quickly. People often waste time searching each system separately or re-asking the same questions. Our project creates a single place to ask a question and get one clear, trusted answer with citations from across all sources. By connecting these tools together, it reduces confusion, saves time, and ensures teams always have access to the most accurate and up-to-date information.

What it does

OneSource is a lightweight “answer engine” that searches your team’s sources (Slack, Google Drive, and GitHub) in parallel, ranks what it finds, and returns a single, cited answer with confidence and freshness. The backend (FastAPI + Postgres) handles OAuth connections, securely stores tokens, and queries each provider concurrently with short timeouts and a tiny TTL cache. A simple Fusion ranker scores candidates (freshness, authority, specificity) and a Policy guard redacts secrets and flags conflicts; every response includes citations and an X-Trace-Id so you can open a full trace showing provider timings, scores, and reasons. The frontend (Vite + React/TS) provides an Ask page (answer + citations/badges), a Connections page (authorize Slack/Drive/GitHub), and a Trace view for transparency. The system is designed to be deterministic and fast for demos, and optionally supports a Gemini-powered rerank/conflict check without changing the public API.

How we built it

Using TS + React Front end Using Python + FASTAPI + postgresql for Back End

Challenges we ran into

A lot of bugs and difficulties with API requests and authentication

Accomplishments that we're proud of

Getting a working product

What we learned

A lot about APIs , A2A and ADK, and authentication

What's next for OneSource

Implementing A2A and ADK (fell short on time)

Built With

• fastapi
• postgresql
• python
• react
• ts