Nebius Personal AI — Hackathon Submission (Nebius x NVIDIA)

Brief overview

This project is a production-oriented Personal AI assistant built for the Personal AI Track. It runs a FastAPI backend that talks to NVIDIA Nemotron models via Nebius Token Factory (or Nebius AI Cloud). It stores short-term persistent memory per user, supports streaming responses for a low-latency UI, and is packaged so you can deploy on Nebius Serverless Endpoints or your own Kubernetes cluster.

Why this project (short, human):

• I chose the Personal AI Track because it showcases ownership of private data, persistent memory, and practical utility: an assistant that keeps context across sessions, runs on open infrastructure, and uses NVIDIA open-source models.
• The system is engineered for judges: it uses Nemotron models via Nebius, provides a responsive streaming UI, and includes production-ready deployment artifacts.

What’s included (high level)

• Backend: FastAPI app (app/main.py) — query and streaming endpoints.
• Model adapter: app/model_client.py — resilient HTTP client with streaming support.
• Prompting + safety: app/prompting.py — light PII checks and prompt builder.
• Persistence: app/db.py — async SQLAlchemy models (SQLite for local, switchable to Postgres).
• Minimal frontend: app/frontend (static HTML/CSS/vanilla JS) with streaming UI.
• Deployment assets: Dockerfile, docker-compose.yml, Kubernetes manifests, Nebius manifest template.
• Tests: tests/test_api.py with pytest + pytest-asyncio + httpx AsyncClient.
• This README and usage instructions for local testing and demo.

Folder structure

See FOLDER_STRUCTURE.txt for a full tree. Key items:

• app/main.py — FastAPI application with /api/v1/query and /api/v1/stream_query (SSE).
• app/model_client.py — HTTP + streaming adapter to Nebius model endpoints.
• app/prompting.py — helper to build prompts and minimal safety checks.
• app/frontend — static demo UI (no Node.js).
• tests — automated tests.

Prerequisites (local)

• Python 3.10+ (3.11 recommended)
• Git, Docker (optional, recommended for local parity)
• (Optional) A Nebius API key for live model calls

Built With

• api
• css3
• docker
• fastapi
• html5
• javascript
• kubernetes
• python
• sqlite