markdown# Meetric — AI Meeting Intelligence

Global AI Hackathon with Qwen Cloud — Track 4: Autopilot Agent

Meetric turns raw meeting transcripts into structured intelligence and then acts on it — drafting follow-ups, detecting blockers, planning sprints, and dispatching messages to email, Slack, and Notion, all with a human-in-the-loop approval checkpoint before anything is sent.

---

What it does

Paste or upload a meeting transcript. Meetric runs a 6-agent pipeline on Qwen Cloud and produces:

Output	Agent
Summary, decisions, action items, participants	Agent 1 — Analysis (+ cross-session memory)
Blocker detection with severity + owner	Agent 2 — Blocker
Sprint plan with story points + velocity	Agent 3 — Sprint Planner
Dependency workflow as a DAG	Agent 4 — Workflow
Prioritised action plan with phases	Agent 5 — Action Plan
Drafted follow-up messages per recipient	Agent 6 — Follow-up

Results appear in a 9-tab dashboard. An inline AI chat lets you edit any field — changes are patched back to the database in real time.

---

Autopilot features (Track 4)

Human-in-the-loop tool dispatch

Every follow-up draft has a Review & Send button. Clicking it opens an approval modal showing exactly which tool will be called and what data it will send. Only after clicking Confirm & Send does the agent execute the tool call. This is the HITL checkpoint required by Track 4.

MCP tool layer (lib/agents/tools.ts)

After approval, the LLM runs an agentic tool-call loop using Qwen's function-calling API:

Tool	Provider	Credentials
send_email	Resend API	App-level RESEND_API_KEY
create_calendar_event	Google Calendar API	Per-user OAuth2 (stored in Supabase)
post_slack	Slack Incoming Webhook	App-level SLACK_WEBHOOK_URL
create_notion_page	Notion API	Per-user integration token (stored in Supabase)

For High-urgency follow-ups the agent automatically books a calendar check-in in addition to sending the message — no extra instruction needed.

Cross-session memory (lib/agents/memory.ts)

Before each analysis run, Meetric fetches the user's last 5 meetings from Supabase and compresses them into a structured context block injected into Agent 1's system prompt. The agent can surface patterns like:

"This is the third sprint in a row where the notification service has been flagged as a blocker owned by Priya."

---

Architecture

┌─────────────────────────────────────────────────────────────────┐ │ Client │ │ Upload zone → Results dashboard (9 tabs + AI chat) │ │ ↓ HITL approval modal │ ├─────────────────────────────────────────────────────────────────┤ │ API layer (Next.js App Router) │ │ /api/analyze /api/chat /api/send-followup /api/memory │ ├─────────────────────────────────────────────────────────────────┤ │ Orchestrator (lib/agents/orchestrator.ts) │ │ Agent 1 (serial) → Agents 2-6 (parallel, Promise.allSettled) │ │ Analysis Blocker Sprint Workflow ActionPlan FollowUp │ ├─────────────────────────────────────────────────────────────────┤ │ Tools / Infrastructure │ │ Qwen Cloud (qwen-plus / qwen-max) Supabase (PostgreSQL + RLS)│ │ Resend (email) Google Calendar Slack Notion │ │ Clerk (auth) │ └─────────────────────────────────────────────────────────────────┘

See public/architecture.svg for the full diagram.

---

Tech stack

• Framework: Next.js 15 (App Router, server components)
• AI: Qwen Cloud — qwen-plus (free tier) / qwen-max (paid), via dashscope-intl API
• Database: Supabase (PostgreSQL + Row Level Security)
• Auth: Clerk
• Email: Resend
• Language: TypeScript

---

Getting started

1. Clone and install

git clone https://github.com/howelldevs/Meetric
cd Meetric
pnpm install

2. Environment variables

Copy .env.example to .env.local and fill in the values:

cp .env.example .env.local

3. Database setup

Run the migrations in your Supabase SQL editor:

# Run in order:
supabase/migrations/20250616_user_integrations.sql

Or paste the SQL directly into the Supabase dashboard → SQL Editor.

4. Run

pnpm dev

Open http://localhost:3000.

---

Connecting integrations

Go to /integrations after signing in:

• Google Calendar — click Connect, approve the OAuth screen. Requires GOOGLE_CLIENT_ID + GOOGLE_CLIENT_SECRET and the callback URI https://yourdomain.com/api/integrations/callback added to your Google Cloud OAuth app.
• Notion — create an integration at notion.so/my-integrations, copy the secret_ token, paste it in the Integrations page.
• Slack — create an Incoming Webhook at api.slack.com/apps, set SLACK_WEBHOOK_URL.
• Email — verify your domain at resend.com/domains, set RESEND_API_KEY + EMAIL_FROM.

All credentials are dry-run safe — if a key is missing, the tool logs the action and returns success so demos work without live credentials.

---

Alibaba Cloud deployment

This project runs on Alibaba Cloud via the Qwen Cloud API (dashscope-intl.aliyuncs.com). See lib/ai-config.ts for the API configuration.

Proof of deployment: see the screen recording in the submission for the backend running on Alibaba Cloud infrastructure.

---

Project structure

lib/ ai-config.ts # Qwen Cloud model config (free / paid tier swap) storage.ts # Supabase admin client, all DB operations agents/ orchestrator.ts # 6-agent pipeline tools.ts # MCP tool layer (email, calendar, Slack, Notion) memory.ts # Cross-session meeting memory app/ api/ analyze/ # Agentic pipeline route chat/ # AI chat + patch persistence send-followup/ # HITL tool dispatch memory/ # Cross-session memory endpoint integrations/ # OAuth connect, callback, Notion token, status results/[id]/ # Results dashboard page integrations/ # Integrations settings page upload/ # Transcript upload page components/ results/ results-dashboard # 9-tab results UI follow-ups # HITL approval flow analysis-chat # Inline AI chat panel supabase/ migrations/ # SQL schema files public/ architecture.svg # System architecture diagram

---

License

MIT — see LICENSE