Skip to content

README.md

Latest commit

 

History

History
261 lines (182 loc) · 8.36 KB

README.md

File metadata and controls

261 lines (182 loc) · 8.36 KB

🧠 AI Meeting Intelligence Copilot

agentkit-challenge Built with Lamatic Next.js Slack Vercel

Turn raw meeting transcripts into structured, actionable intelligence — automatically delivered to Slack.

Built by Vijayshree Vaibhav.

🎯 Problem Statement

Teams walk out of meetings with scattered notes, untracked action items, and unclear ownership. Critical insights get lost in long conversation threads. Following up manually is slow and inconsistent.

This agent solves it in one step: paste a meeting transcript → receive a structured summary, action items, risks, next steps, and a follow-up email draft — all delivered to Slack automatically.

✨ What It Does

Input Output
Raw meeting transcript Executive summary
Any language, any format Action items with owners
Paste or type directly Risks & blockers
Suggested next steps
Follow-up email draft
Slack notification ⚡

🏗️ Architecture

graph TD
    A[👤 User] -->|Pastes transcript| B[Next.js Frontend\nkits/embed/chat]
    B -->|Sends message| C[Lamatic Chat Widget\nchat-v2]
    C -->|POST /widget/chat| D[Chat Trigger Node\nLamatic Studio]

    D --> E[🧠 Generate Text\nLLM - Executive Summary\nAction Items · Risks · Next Steps\nFollow-up Email]
    E --> F[📋 Generate JSON\nStructured Output]

    F --> G[⚡ Slack API Node\nWebhook POST]
    F --> H[💬 Chat Response\nStreamed back to widget]

    G -->|Formatted insight card| I[📨 Slack Channel]
    H -->|Rendered markdown| B

    style A fill:#6366f1,color:#fff
    style B fill:#1e293b,color:#fff
    style C fill:#1e293b,color:#fff
    style D fill:#0f766e,color:#fff
    style E fill:#7c3aed,color:#fff
    style F fill:#7c3aed,color:#fff
    style G fill:#4A154B,color:#fff
    style H fill:#0f766e,color:#fff
    style I fill:#4A154B,color:#fff
Loading

Flow inside Lamatic Studio:

Chat Widget ──▶ Generate Text (LLM) ──▶ Generate JSON
                                               │
                              ┌────────────────┤
                              ▼                ▼
                          Slack API      Chat Response
                         (Webhook)     (back to user)

📸 Screenshots

Live Web App — Transcript Input & Structured Output

Web App Running

Lamatic Studio — Flow Execution & Preview

Lamatic Flow Running

Slack Integration — Auto-delivered Insights

Slack Integration

Frontend Landing Page

Landing Page

🛠️ Tech Stack

Layer Technology
Frontend Next.js 14 (App Router)
UI Tailwind CSS + shadcn/ui
AI Workflow Lamatic Studio
Chat Widget Lamatic chat-v2 embedded widget
LLM Configured in Lamatic (GPT-4o / Claude)
Integrations Slack Incoming Webhooks
Deployment Vercel

⚙️ Quick Start

1. Clone

git clone https://github.com/vijayshreepathak/AgentKit.git
cd AgentKit/kits/embed/chat

2. Install

npm install

3. Configure environment

Create .env.local inside kits/embed/chat/:

NEXT_PUBLIC_LAMATIC_PROJECT_ID=your_project_id
NEXT_PUBLIC_LAMATIC_FLOW_ID=your_flow_id
NEXT_PUBLIC_LAMATIC_API_URL=https://your-project.lamatic.dev

Get these values from your Lamatic Studio project settings.

4. Run

npm run dev
# → http://localhost:3000

🔧 Lamatic Studio Setup

Flow Structure

Build this flow in Lamatic Studio:

[Chat Trigger] → [Generate Text] → [Generate JSON] → [Slack API]
                                                    → [Chat Response]

Chat Trigger Configuration

  1. Open your flow → click the Chat Trigger node
  2. Under Allowed Domains, add * (for development) or your production domain
  3. Click Save → Deploy

Generate Text Prompt (LLM Node)

You are an AI Meeting Intelligence Copilot. Analyze this meeting transcript and produce:

## Executive Summary
[2-3 sentence summary of the meeting]

## Action Items
- **[Owner]:** [Task] — due [date if mentioned]

## Risks & Blockers
- [Risk or blocker identified]

## Next Steps
- [Recommended follow-up action]

## Follow-up Email Draft
Subject: [Meeting Topic] — Decisions & Next Steps
[Professional email body]

Meeting transcript:
{{trigger.chatMessage}}

Slack Webhook

  1. Create an Incoming Webhook in your Slack workspace
  2. Add the webhook URL to the API Node in Lamatic
  3. The agent automatically sends the structured output to your channel

🚀 Deploy to Vercel

Deploy with Vercel

After clicking Deploy:

  1. Fill in the three env vars from your Lamatic Studio
  2. Vercel builds and deploys automatically
  3. Add your Vercel production URL to the Chat Trigger's allowed domains in Lamatic

📁 Project Structure

kits/embed/chat/
├── app/
│   ├── page.js                    # Landing page (Server Component)
│   ├── layout.js                  # Root layout with fonts + analytics
│   ├── globals.css                # Tailwind v4 + CSS variables
│   └── Screenshots/               # Demo screenshots
├── components/
│   ├── LamaticChat.js             # Lamatic widget lifecycle manager
│   ├── HeroActions.jsx            # CTA buttons (Client Component)
│   ├── TranscriptPlayground.jsx   # Transcript input + analyze flow
│   └── ui/                        # shadcn/ui components
├── flows/
│   └── embedded-chatbot-chatbot/  # Exported Lamatic flow config
├── .env.local                     # Local env vars (not committed)
└── package.json

💡 Key Engineering Decisions

Why data-* attributes on the root div? The Lamatic chat-v2 widget reads data-api-url, data-flow-id, and data-project-id from #lamatic-chat-root at initialization. These must be set before the script runs.

Why bootstrap on mount (not on button click)? The widget fetches chatConfig and creates an IndexedDB session on init. Bootstrapping early means the session is ready before the user sends their first message — preventing "unexpected error" on first send.

Why not remove the root on unmount? The Lamatic widget uses IndexedDB with the session key as a primary key. Removing + re-adding the root causes a ConstraintError: Key already exists because the old session record is still in the database. The root is intentionally a page-lifetime singleton.

🔮 Future Improvements

  • Audio/video transcript upload (Whisper transcription)
  • Multi-language meeting support
  • Meeting history dashboard
  • Calendar integration (Google Meet / Zoom auto-fetch)
  • Team analytics on action item completion rates
  • Export to Notion / Confluence

👩‍💻 Author

Vijayshree Vaibhav

LinkedIn GitHub

📜 License

MIT — see LICENSE.