Hacker
Mohith
@mohiths
Projects
Take Two
Flip the Switch HackathonNov 2026
Take Two keeps a film shoot day in sync when plans change. On an indie set the plan breaks all day: a location refuses, an actor is late, a recorder dies, a scene runs over. Today the AD rewrites the schedule by hand, re-posts it in chat, and nobody is sure who has seen the new version. Take Two is a production room built on Switch. When something changes, it holds the affected scenes, asks two agents for help (a Script Supervisor on Claude Code says what each scene must keep for the story; a 1st AD on Codex drafts up to three concrete plans), checks every plan in code against hard facts (cast hard stops, kit, locations, time), and lets the director pick on a native Switch question card with a rendered set model of each option. The chosen plan goes out as a numbered, colour-coded revision with a call sheet to the production room (Mattermost) and the crew room (Discord), and Take Two tracks who has confirmed their new duties. A card asked on old facts closes itself, so nobody approves a stale plan. Agents propose, code checks, people decide. It is for small film crews, student productions and indie directors who run a shoot day from a group chat.
- Top 50
Arcloom: Evidence That Survives Agent Handoffs
Graph HacksOct 2026
Security investigations often lose context when a session ends or another investigator takes over. A useful finding gets buried in a report, and its connection to the next investigation disappears. Arcloom turns reviewed security evidence into persistent, connected memory. A researcher submits a report, a human approves a bounded test, and a separate review determines whether its evidence becomes a confirmed finding. FalkorDB connects those findings to the capabilities they enable and the tests that require them. A fresh investigator can then propose a related test with an inspectable supporting path, without inheriting the previous conversation. Our demonstration follows a gateway finding into a separate invoice investigation and then a chain requiring both findings. It also preserves negative results, researcher attribution and deployment history. Every new execution requires approval, and a deployment change prevents outdated findings from silently authorizing new work. Built for security researchers, reviewers and teams developing accountable agent workflows.
Into the unknown@sansynx