.png&w=256&q=75)
🤓 Latest Submissions
Fieldwork: Voice User Interviews at Scale
Talking to users is the most useful thing a founder can do, and the first thing to get skipped. Interviews take hours to schedule, run and write up, and the insight ends up buried in fifty transcripts. Fieldwork automates that. A founder writes a research goal and a few seed questions, then shares one link. Anyone who opens it has a real voice conversation with an AI interviewer, with no account or install. It asks one question at a time, follows up on specifics ("walk me through the last time that happened"), stops the moment you talk over it, and wraps up by itself once the research goal is covered. After each call, an async pipeline pulls out takeaways, sentiment and verbatim quotes. One click synthesizes up to five themes ranked by how many interviews raised them. Every quote is checked in code against what the participant actually said and links to the exact line of its transcript. AssemblyAI Universal-3 Pro Streaming drives the conversation. Its end_of_turn signal decides when the participant has finished, tuned for people who pause to think. Partial transcripts power barge-in and live captions. Replies stream sentence by sentence into Amazon Polly, so the first audio arrives about 1-2 seconds after the participant stops talking. Under the hood it's three planes: a FastAPI orchestrator on ECS Fargate holding the live WebSocket, a serverless API for the dashboard, and SQS + Lambda workers for analysis with retries and dead-letter queues. All of it is Terraform. Open models on Featherless were picked by benchmarking 24 of them.
30 Sep 2026
