# Context Desk

- **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:** [sayam ajmal](https://cerebralvalley.ai/u/SaneSayam)
- **GitHub:** https://github.com/SayamDev/context-desk
- **Demo video:** https://youtu.be/XepNALHx4Z0
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4110

Context Desk is an AI clip farming studio. It turns live streams and long episodes into story-complete, subtitled Shorts in any language, on the editor's own computer.

The problem: clip channels and small creators spend hours scrubbing long streams and episodes for moments worth posting, then subtitling them, often from a language they don't speak. Most auto-clippers cut random high-energy seconds and machine-translate them, so the clip loses the story and the meaning.

What it does:
- Paste a link, upload a video, or point it at a YouTube, Twitch or Kick live stream.
- Meta Muse Spark reads sampled frames and the transcript and finds complete moments: a setup, a turn and a payoff. Every clip gets an English title, a reason and its place in the source.
- Muse Voice transcribes speech in 25 spoken languages. Muse Spark writes natural, dialect-aware subtitles and flags every line it is unsure about, so the editor checks the right words.
- Culture notes explain slang and references on screen.
- One click reframes to 9:16 with face tracking. It adds caption styles, title cards and a post kit (caption, hashtags, notes).
- An optional English dub clones each speaker from their own line.
- For the whole episode it writes chapters, answers questions (Ask the episode) and renders a one-minute trailer.
- Live clipping records shared streams at up to 1080p in the background, with a monitor window and an alert for every new clip.

In our test, a 4-minute episode excerpt gave 4 ready clips in about 2 minutes, and 125 English subtitle lines for the excerpt in about 2.5 minutes.

Everything was built during the hackathon: the repository's first commit is 31 Oct 2026, and all 123 commits fall between 31 Oct and 4 Nov. The code is open source (MIT).

## More from Meta Global AI Developer Hackathon

- [HotBot](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4107)
- [Kore Meta](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4108)
- [Easel](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4109)
- [Team Abrar](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4111)
- [Ernest Tretjakov](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4112)
- [DROQ](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4113)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4110. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt