AI Products / 2025–26
Actual Voice
Actual Voice collects open-ended spoken feedback and produces group-level themes, engagement signals, and summaries while keeping raw individual responses out of aggregate reports.
What Actual Voice does
Actual Voice lets people answer a question by speaking. The product transcribes each response, analyzes it, and combines the results into recurring themes, engagement signals, and a group summary.
The processing path has several stages: upload, storage, transcription, per-response analysis, group aggregation, and summary generation. Each stage records its status so failed work can be retried and production issues can be traced.
Processing pipeline
Uploads enter a tenant-aware pipeline backed by Supabase Storage, Postgres, Edge Functions, and worker queues. Each stage records enough state to retry safely. Debounced aggregation prevents every new response from triggering redundant group computation, while fallback handling keeps temporary queue or model failures from blocking the respondent experience.
Per-response analysis stores relevance information before group aggregation. This filters weak or off-question themes early and gives the summary step cleaner inputs.
Summary validation
Runtime checks enforce the summary’s privacy boundary, required structure, and length. Production validation regenerates insights from the current data so a review does not accidentally measure a cached result.
The pipeline supports monitoring, retries, backfills, and regression tests using real questions and privacy-safe transcript data.