Uxie

Inspiration

The CSSS webpage's IDE-style look inspired the interface, so Uxie looks and feels like a developer tool, not a generic dashboard.

What it does

The problem behind it is that when you open an unfamiliar codebase, it's hard to know what depends on what, and changes break things you didn't expect.

How we built it

Four people each built one stage of a pipeline: Indexer (C++): uses Tree-sitter to parse the repo into functions, imports and calls, and writes them to SQLite. Embeddings (Python): embeds each symbol's source and supports semantic search over them with numpy. Server (Python, FastAPI): loads the graph, builds the prompt and streams the LLM's answer over a WebSocket. Front end (React, TypeScript): React Flow layout that communicates frequently over the backend.

Challenges we ran into

• The initial idea
• Connecting the initial pipeline
• Project Management (Git)

Accomplishments that we're proud of

A full working pipeline from repo to graph to AI answer in one hackathon

What we learned

Agreeing on everything early, and planning in depth before starting matters more than anything else.

What's next for Uxie

• More language support
• Planned VR & ElevenLabs

Built With

• cmake
• cpp
• python
• react
• reactflow
• typescript