AI investigative assistant for the NC State Hackathon 2026. Takes mixed evidence (video + text) and reconstructs what happened using TwelveLabs, Gemini, and ElevenLabs.

Theme: Urban Noir — crime, mystery, surveillance, deception, fragmented truth.

• Video analysis — Extract key events with timestamps using TwelveLabs
• Evidence fusion — Combine video + text into structured JSON via Gemini
• Noir narration — Generate detective-style audio summary with ElevenLabs

NightArchivist/
├── backend/          # FastAPI (Python)
├── frontend/         # Next.js (TypeScript)
├── data/             # Local case storage (created at runtime)
└── README.md

• Python 3.10+
• Node.js 18+
• API keys: Gemini, TwelveLabs, ElevenLabs

Create backend/.env from the example:

cp backend/.env.example backend/.env

Edit backend/.env and add your API keys:

GEMINI_API_KEY=your_gemini_api_key
TWELVELABS_API_KEY=your_twelvelabs_api_key
ELEVENLABS_API_KEY=your_elevenlabs_api_key

Where to get keys:

• Gemini API
• TwelveLabs — sign up and create an API key
• ElevenLabs — sign up and create an API key

TwelveLabs index (optional): If you have an existing index ID, set TWELVELABS_INDEX_ID in .env. Otherwise the backend will create one at first use (index creation can take a minute).

cd backend
pip install -r requirements.txt
uvicorn main:app --reload --host 0.0.0.0 --port 8000

The API runs at http://localhost:8000.

cd frontend
npm install
npm run dev

The app runs at http://localhost:3000.

If the backend is on a different host/port, create frontend/.env.local:

NEXT_PUBLIC_API_URL=http://localhost:8000

Full demo (no APIs): Set DEMO_MODE=true — skips TwelveLabs, LLM, and ElevenLabs. Uses mock data for everything.

Hybrid (real video + audio, mock LLM): Set MOCK_LLM_ONLY=true — TwelveLabs and ElevenLabs run for real; only the reasoning/fusion step (Gemini/LLM) is mocked. Use when LLM APIs are failing but TwelveLabs and ElevenLabs keys work.

1. Upload — Select a video file (mp4, mov, etc.) and a text file (.txt) with evidence
2. Processing — Wait while the pipeline runs: Video Analysis → Reasoning → Narration
3. Results — View case summary, timeline, contradictions, open questions, and play the noir narration

MIT