# Neural Nexas

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Nayananshu Garai](https://cerebralvalley.ai/u/NG007)
- **GitHub:** https://github.com/N-Garai/Chronocut
- **Demo video:** https://youtu.be/BodSlfCz9PM
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2216

ChronoCut is an agentic AI video editor that turns raw footage into a finished, rendered cut — automatically.

The problem: Creators sit on hours of raw footage — vlogs, event coverage, product clips — and editing it is the slowest, most expensive step of publishing. Consumer template apps force your video into preset formats, and professional editors are out of reach for most creators. Nobody has an "editor-in-a-box" that actually watches your footage, writes the story, and makes real editing decisions.

What ChronoCut does: You upload media (or paste a prompt or a reference link), and a governed multi-agent pipeline does the work of an editing team:
- Archivist transcribes speech and tags your photos/videos (multimodal understanding of the actual material),
- Screenwriter stage builds a narrative script and 3-act arc,
- Director composes the shot list and issues edit decisions (cuts, grading, motion, sound design) grounded in skill files,
- Editor supervisor refines the plan with six batched specialist sub-agents — rhythmist (micro-trims), motionist, colorist, transitionist, audioist, typographer — then assembles a machine-validated Edit Decision List,
- Everything renders to a real downloadable MP4 via ffmpeg, with an OpenTimelineIO export for pro NLE handoff.

After the first cut, you stay in a timeline editor and chat with the agents in plain English ("tighten the middle section, add captions") — edits apply as validated operations with instant undo, and every agent action streams into a live telemetry feed.

Under the hood: a model-governance seam (cache → rate limiting → per-job cost budget → in-flight coalescing → automatic failover chain) keeps every call reliable and cost-capped, and honest telemetry shows the provider and model behind every response. The whole stack runs free: FastAPI API on Render, static editor on Cloudflare Pages, GPU-free rendering on GitHub Actions, Firestore for state, Supabase for storage.

## More from Meta Global AI Developer Hackathon

- [Vinklings](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2213)
- [Capin](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2214)
- [BolSaakshi](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2215)
- [OhnoTrash](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2217)
- [MetaHackers](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2218)
- [CallOrder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2219)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2216. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt