inferred from GitHub
timbre
Timbre: two agents fight so your voice survives. Gemini multi-agent pipeline that writes in a founder's voice and verifies facts. Google I/O Hackathon (Cerebral Valley), May 2026.
Core trick
Gemini agents write drafts in a founder's voice, then a separate verifier agent checks each factual claim against research sources before publishing.
Evidence
README matched ""Cerebral Valley" "Opus 4.7"": README.md
README excerpt
<div align="center"> # Timbre **Two agents fight so your voice survives.** A multi-agent content engine for technical founders. Researches what matters, writes in your voice, and verifies it didn't lie to sound better. *Hackathon project — Cerebral Valley × Google DeepMind I/O Hackathon, San Francisco, 2026-05-23. Reference build.* [**Live landing → usetimbre.ai**](https://usetimbre.ai) · [**Live dashboard → usetimbre.ai/app**](https://usetimbre.ai/app) · [Architecture](#architecture) · [Specs](specs/) · [Submission](SUBMISSION.md) </div> --- > **Status (Oct 2026):** the landing page and cached-replay dashboard are static and still served; the live API backend ran on a home machine for the event and is offline, so live Scout/Research runs from the hosted dashboard no longer work. Run locally to exercise the live path. ## The bet Technical content is the highest-ROI acquisition channel for B2B founders — and the channel AI categorically fails at. Generic LLM output is detectable. Outsourced copy doesn't carry the founder's voice. Style-transfer tools silently mangle technical facts to sound smoother. Timbre's bet: **voice preservation, not generation quality, is the missing primitive.** Build the engine the founder needs — one that fights to keep their voice through a long agentic pipeline. ## The demo moment After Research streams its reasoning live and Write drafts an article, Voice rewrites it sentence-by-sentence