Back to portfolio
ProductIn production2025–Present

AirLog

Voice-first user testing platform. Testers speak feedback during sessions; AirLog transcribes with self-hosted Whisper, classifies and summarizes with OpenAI, then ships stakeholder-ready reports. Built end-to-end as a multi-tenant dashboard + transcription pipeline + GitHub PR → scene sync — so teams capture durable feedback without brittle typed notes or slow manual QA synthesis.

Gallery

Voice capture

One-click browser recording with pause/resume — speak naturally while testing, ~5× faster than typing detailed notes.

AI classify & summarize

Notes auto-tagged as Bug, Feature, UX, Performance, or Other — plus session and per-note AI summaries.

Report & share

PDF exports, public report tokens, and email delivery so stakeholders get insights without digging through raw notes.

Product

AirLog is a voice-based user testing platform built for teams that need structured, bias-resistant feedback — not scattered spreadsheets. Admins create sessions and scenes, testers join with a code and record as they go, and AI turns raw audio into classified notes, analytics, and shareable reports. Multi-tenant companies, teams, and GitHub PR sync keep the loop tied to real shipping work.

Problem

Typed notes lose context mid-test. Feedback lands in spreadsheets and Slack. Manual tagging and report writing burn hours. Testers influence each other when everyone can see live notes. Merged PRs don’t automatically become “what to test.”

Solution

Browser voice capture → self-hosted Whisper queue → AI classification & summaries → analytics + PDF/email reports. Isolated feedback during active sessions; full reveal after. Multi-tenant RBAC for companies/teams, plus GitHub OAuth that turns merged PRs into test scenes (with backlog for what didn’t make the window).

Highlights

  • Voice recording + self-hosted faster-whisper (Docker/Fly) with retry queue, editable transcripts, and original audio playback
  • OpenAI auto-classification (Bug / Feature / UX / Performance / Other) and session / per-note summaries
  • Multi-tenant company system with RBAC: super admin → company owner/admin → tester
  • Bias prevention: testers only see their own notes live; post-session reveal for admins
  • GitHub OAuth + PR → scene sync (AI “what to test” extraction, backlog, CI webhook API)
  • Analytics: category, temporal, engagement, historical comparison (Recharts)
  • Stakeholder reports: @react-pdf/renderer PDFs, shareable public tokens, email via SMTP/Resend
  • Installable PWA, realtime notifications (Supabase), audit/usage monitoring for Whisper / OpenAI / SMTP

Specs

Type
AI product / UX research tooling
Role
Founder · Full-stack engineer
Company
Independent
Live
airlog-pro.vercel.app
Focus
Voice feedback, AI analysis, session reporting
Period
2025–Present
Stack core
Next.js 14 · Supabase · Whisper · OpenAI
Auth
Three-tier JWT (jose) · OTP email verification
Traction
Adopted by my employer and now integrated into their product suite

How it works

  1. 1Create a session — Define scenes/tasks, schedule windows, invite teams or share a join code.
  2. 2Testers record — One-click voice (or text) notes per scene while testing; feedback stays isolated.
  3. 3Transcribe & classify — Whisper queue produces transcripts; OpenAI tags categories and can summarize.
  4. 4Analyze & ship — Admins review analytics, generate AI session summaries, export PDF / share / email reports. Optional: sync merged GitHub PRs into scenes for the next round.

Achievements

  • Adopted by the company I was working at and integrated into their product suite as an ongoing part of their offering
  • Shipped a production multi-tenant app: admin, company, and tester surfaces on one Next.js codebase
  • Self-hosted Whisper service (faster-whisper, VAD, QA-tuned prompts) with status tracking, retries, and usage monitoring
  • Full feedback pipeline: record → store (Supabase Storage) → transcribe → classify → summarize → report
  • GitHub integration hub: OAuth, repo linking, PR→scene matching by schedule, backlog, and external sync API + GitHub Action
  • Role privileges, soft-delete users, company invites / join-request workflow, and subscription-style limits
  • Bias-aware session design plus poll/survey support alongside freeform voice notes
  • PWA-ready tester experience with dark/light UI (shadcn/ui + Tailwind) and Three.js landing moments

Stack

TypeScriptNext.js 14ReactTailwindshadcn/uiSupabasejosebcryptjsOpenAI APIfaster-whisperGitHub OAuth / APINodemailerResend@react-pdf/rendererRecharts@dnd-kitThree.js / R3FVercelFly.ioPWA
Core StackFrontendBackend & DatabasesCloud & DevOpsData & Machine LearningDesign & DX

Links

Live product