Impact
0
third-party servers touched
100%
auto-captured meetings
0min
manual note-taking time
v1.0
live in production
The Problem
Manual notes & a data privacy crisis
Empowered Pensions advisors were manually taking notes during every Google Meet call with clients — losing detail, spending time post-call reconstructing conversations, and creating inconsistency across the team.
The obvious fix — public transcription tools like Otter.ai or Fireflies — was off the table entirely. Empowered Pensions handles highly confidential pension and financial data for UK clients. Routing any conversation through a third-party AI transcription service would violate data governance obligations and client trust.
The constraint: Any solution had to keep transcript data entirely within Empowered Pensions' own infrastructure. No audio or text could leave to an external vendor's servers under any circumstances.
Evidence
The extension in action
Extension installed — ID: gcpjnjpmbhegkibeidkabhcjdkacjmmo
Auto mode records every Google Meet automatically. Manual mode lets staff choose when to start.
In-meeting banner confirms active capture — "Do not turn off captions"
Meeting history dashboard — webhook delivery status, download, and replay per session
Architecture & Solution
How the transcript pipeline works
01
Chrome Extension · Content Script
Inject into Google Meet
The extension injects a content script into any
meet.google.com tab. On meeting join, it auto-enables captions if running in Auto mode — no action needed from the advisor.02
DOM Observer · Real-time Capture
Capture captions from the DOM
A MutationObserver watches Google Meet's live caption DOM nodes. Every caption segment is captured as it appears — speaker, text, and timestamp — building a structured transcript in memory.
03
Local Storage · Chrome Extension API
Store transcript locally
The full transcript is saved to Chrome's local extension storage as a structured text file. No data leaves the browser during the meeting. The meeting history dashboard shows all 10 most recent sessions.
04
Webhook · POST on Meeting End
Fire webhook to client database
When the meeting ends, the extension fires a webhook POST carrying the full transcript directly to Empowered Pensions' own database endpoint. Status is tracked per session —
New, delivered, or failed — visible in the meeting history view.05
Client Database · Private Infrastructure
Transcript stored in Empowered Pensions' DB
The transcript lands in Empowered Pensions' own database — fully under their control, accessible for client records, meeting notes, and compliance review. Zero copies on any external server.
Implementation Decisions
Why it's built this way
01
Caption scraping, not audio recording — capturing Google Meet's own live captions via DOM observation means no audio is ever processed or stored. This is both simpler and more privacy-compliant than audio-based transcription.
02
Auto mode by default — advisors forget to start recordings. Auto mode triggers on every Meet join so no session is ever missed without a deliberate manual override.
03
Webhook fires on meeting end, not during — batching the full transcript into a single POST on session close keeps the pipeline simple, avoids partial writes, and reduces database load compared to streaming line-by-line.
04
Webhook status tracked per session — the meeting history dashboard shows delivery status (New / delivered / failed) so staff can manually re-fire a failed webhook without losing the transcript.