"Reticet team on AssemblyAI - Voice Agent Hackathon Hackathon"

Team Idea

Reticet: speech to a case record where nothing is written that was not said. A protection officer finishes an interview in a displacement setting and has minutes to write it up before the next one. That note is what triggers a referral, or fails to. Voice tools built for meetings transcribe and summarise it confidently, including the parts they got wrong. A confident wrong summary is worse than no summary, because it gets acted on and an empty field gets checked. Reticet takes the write-up, not the interview. No survivor is ever recorded. A field worker speaks a case note in English or French, and every populated field has to point at the exact words they said. If those words are not in the transcript, the field stays blank with the reason showing. When the worker says "names I did not get" or "I'm not certain", the system may not fill that field at all. Built on AssemblyAI streaming speech to text. The claim is calibrated abstention.

Team Members

ElkeDikoume

elkedikoume713

Adjaho Nathan

nadjaho

Submission

Reticet AI: a voice agent that knows who said it

A voice debrief agent for anti-trafficking caseworkers. Code saves a proposed field only when the caseworker’s words support it, it is read aloud, and the caseworker confirms it. “I think” is labeled as inference, not evidence.