The shortest distance
between two languages.

1. 01

   Accuracy first

   Dual-agent verification, event- and booth-level context, and term/mishearing dictionaries mined from real meetings injected into the next event recognizer. We measure error rates per accent (Singaporean, Filipino, Malaysian English) against offline high-accuracy transcripts.

2. 02

   Co-reviewed by default

   After every meeting, both parties review the summary line-by-line and sign. Nothing leaves the booth without consent.

3. 03

   On-site by default

   During beta our team operates every event on-site. Self-serve opens with the 1.0 release.

4. 04

   Data as moat

   Consented Korean–Southeast Asian English meeting speech is pseudonymised into term/mishearing dictionaries and an accent-labelled speech corpus — 39,921 utterances across 6 events in 3 countries as of Sep 2026. A competitor can license the same models; they cannot get this field data.

Audio processed under required consent is auto-purged after 90 days; only audio with separate voice-corpus consent is kept until withdrawal. Every sub-processor (Soniox, OpenAI, Anthropic, Google, Deepgram, Supabase, Vercel, Resend, Daily.co, Sentry, Cloudflare) operates under a data processing agreement or SCC. Full details in the Privacy Policy.