50 meetings a week. That’s the average for a mid-sized tech team — decisions flying, ideas sparking, but poof: 70% forgotten within a day.
airCloset, Japan’s slick fashion subscription service, just flipped the script. They’ve built an automated meeting intelligence system with Google Meet, Slack, and RAG that captures every word, slide, and decision, then serves it up on demand. No notes. No rewatches. Just ask, and know.
And here’s the kicker — it’s not some sci-fi dream. CTO Ryan Tsuji rolled this out because, as he puts it:
Face-to-face communication is fast and dense. A decision that takes 30 minutes over text can happen in 5 minutes in a meeting. That’s the biggest advantage of meetings.
But that gold? It evaporates. ‘What did we decide?’ ‘Where’s the recording?’ Sound familiar?
Why Bother When Google Already Does Notes?
Look, Google Meet’s got transcription now, even Gemini notes. Cute. But it’s siloed — buried in your calendar, permissions a nightmare, search? Forget it. You won’t scrub through hour-long vids for that one release date.
airCloset’s twist? They pipe everything into Slack — your team’s real nervous system — with instant notifications, auto-permissions for channel folks, and a bot that wields RAG like a memory wizard. Mention the bot: “What was the release date last week?” Boom: answer with sources. Screen shares? Indexed too. It’s meetings as a living database.
Think of it like this: meetings are verbal fireworks — brilliant, fleeting. This system bottles the sparks, labels them, and hands you a searchlight.
One paragraph wonder: Dead simple to start.
Click “AI Fassy Meet” in Google Calendar (their Chrome extension magic). Pick a Slack channel — favorites bubble to the top. Hit create. Boom — Meet link set, recording/transcription/notes all on by default. No fiddling.
Recurring? Last channel pre-selected. Lazy genius.
Meeting ends. Slack pings: “It’s over.” Minutes later: links drop. Mention bot for summary. Structured, cited, screen-shares parsed via Gemini 1.5 Flash. Past meetings? RAG pulls from the vault.
We’re talking embeddings from Vertex AI, Firestore for storage, Pub/Sub for events — a lean Cloud Run stack that feels like butter.
But here’s my unique hot take, the one Tsuji’s post dances around: this isn’t just productivity porn. It’s the email killer for decisions. Remember 1996? Hotmail turned clunky SMTP into instant comms. This? It turns meetings — that sacred cow of office life — into queryable atoms. Bold call: by 2027, every Fortune 500 runs a clone. Async dies; collective brain lives.
How Does RAG Make Meetings Searchable Anyway?
RAG — Retrieval-Augmented Generation — your new best friend. (If you’re new: it’s AI that first digs docs for context, then generates answers. Hallucination-proof-ish.)
Here’s the flow, unvarnished:
-
Meet ends → workspace-pipeline subscribes to events.
-
meet-pipeline grabs recording, transcript, screen shares. Embeds chunks with gemini-embedding-001. Stores in Firestore/BigQuery.
-
Permissions auto-granted: Slack members, invitees, calendar peeps. No admin hell.
-
Slack bot? Hono on Cloud Run. Query hits → retrieve relevant embeddings → Gemini 1.5 Flash summarizes with links.
Screen shares? OCR’d and embedded. Code snippets? Searchable. It’s like giving your meetings a Google.
Tsuji’s team extracted shared logic — Drive access, caching in Redis — into a package. Smart. Scales across bots and extensions.
Analogy time: imagine meetings as a bustling marketplace — vendors yelling deals, crowds swirling. Without this, it’s chaos post-close. With RAG? It’s a indexed ledger, every bargain logged, queryable by keyword.
And the beauty? Open-ish stack. Vertex AI, but swap for open models? Easy. This screams open-source remix.
Slack thread turns epic.
Ends with: notify → links → “Summarize?” → bot delivers.
The Bot searches past meeting transcripts, summarizes the relevant parts, and responds with source links.
Infinite scroll no more.
The Tech Stack That Powers the Future
No monolith here. Modular glory:
-
Chrome Extension (Manifest V3): UI sparkle.
-
Cloud Run APIs: Hono-fast.
-
Pub/Sub → Run: Event firehose.
-
Gemini 1.5 Flash + embeddings: AI muscle.
-
Firestore/Storage/Redis/BigQuery: Data fortress.
Slack Block Kit for those crisp notifications. Workspace APIs for the glue.
Cost? Negligible for most — embeddings cheap, Runs sleep when idle.
Critique corner: Google’s native notes are PR bait — half-baked, no cross-app search. airCloset calls BS by building around it, not on it. Respect.
This system’s the first domino. Digitize ops? Start here. Tsuji’s prior MCP servers (991 tables NLQ’d!) were warm-up; this is the main event.
Picture warehouses of the mind — every convo, indexed eternally. AI’s platform shift? It’s making human bandwidth infinite.
What Happens When Every Company Copies This?
Productivity explodes. Repetitive yak-shaves vanish. “We discussed that Tuesday” becomes “Bot, pull it.”
Downside? Privacy watch — but auto-permissions mitigate. (Still, audit those channels.)
Prediction: Slack/GitHub/Teams plugins inbound. Open-source the extension? Beat goldmine.
Energy high: this feels like 2004 web 2.0 — user-gen content exploding. Now, AI-gen knowledge from spoken word.
🧬 Related Insights
- Read more: 45 Browser Tools That Nuked My Endless Tab Hell
- Read more: The Tests Your Team Skips Until Customers Scream
Frequently Asked Questions
What is an automated meeting intelligence system?
It’s AI that records, transcribes, indexes, and queries meetings automatically — turning video hours into instant Slack answers via RAG.
How do you integrate Google Meet with Slack for transcripts?
Use a Chrome extension for one-click Meets, Workspace Events API for end-of-meeting triggers, then pipe to Slack with auto-links and a RAG bot.
Can I build a RAG search for my own meeting transcripts?
Yes — embed with Gemini or open models, store in vector DB, query via bot. airCloset’s Cloud Run stack is a blueprint.
