Unfiltered

Always-on AI voice interviewer for honest, multimodal user research.

• Synthetic users to pre-flight your interview guide before burning real-participant goodwill
• Real interviews by an AI bot, on demand, any time of day — no scheduling
• Evidence-grounded findings with clip links, quotes, and frames for every claim

Status

Phase 0 — Foundation. The repo was just rebuilt from scratch on a Cloudflare-native stack. Active scaffolding; not yet deployable.

See docs/v2-architecture-spec.md for the full plan, decisions, and phased build schedule.

Tech Stack

• Next.js 15 on Cloudflare Workers (@opennextjs/cloudflare)
• Cloudflare D1 + Drizzle ORM
• Cloudflare Durable Objects (per-session state + WS fanout)
• Cloudflare Workflows (analyzer, synthetic, guide creator)
• Cloudflare R2 (recordings)
• Cloudflare AI Gateway → Gemini, GLM, Claude, OpenAI
• WorkOS (auth)
• Resend (email)
• Vapi (voice runtime)

Getting Started

npm install
cp .env.example .env.local       # fill in development values
wrangler d1 create unfiltered-db # paste id into wrangler.toml
npm run db:generate              # generate migrations from schema
npm run db:migrate:local         # apply to local D1
npm run dev                      # http://localhost:3000

Documentation

• Canonical spec: docs/v2-architecture-spec.md
• Legacy V1 reference (prompts, research notes, old docs): legacy-prompt-specs/
• Codex/agent context: CLAUDE.md