loftus

01 / Challenge

Teams had no passive way to document meetings — someone always had to take notes manually, leading to missed details, inconsistent records, and no searchable history of decisions.

02 / Solution

A headless Playwright worker authenticates with Google, joins scheduled Meet sessions in ghost mode (muted, no camera), records audio via the Web Audio API, and pipes it through faster-whisper for transcription. A Django backend orchestrates session lifecycle, generates structured AI notes via Gemini, and chunks + embeds both transcripts and notes into Supabase with pgvector. A RAG query layer lets users ask natural-language questions over their entire meeting history. n8n webhooks handle automated document delivery, and a Next.js dashboard ties it all together.

03 / Impact

Fully automated end-to-end meeting capture, documentation, and retrieval pipeline. Users can query across all past meetings semantically — no manual note-taking or digging through recordings required.

04 / My Role

Sole author. Designed and built the entire stack — the Playwright bot with ghost-mode join logic and audio capture, the Django API and session orchestration, the faster-whisper transcription pipeline, the RAG indexing and retrieval system using pgvector, and the n8n webhook integration for automated delivery.