Skip to content

patilanupam/Clarify_Meet_AI

BranchesTags

Repository files navigation

ClarifyMeet AI 🤖

Link: https://clarifymeetai-btra6m4p2msf5es6z9rdnc.streamlit.app/

Transform meeting conversations into actionable insights using AI

ClarifyMeet AI is an intelligent meeting minutes generation tool that automatically extracts structured information from meeting transcripts using LangGraph and Ollama. Simply upload a text transcript, and AI will extract summaries, action items, decisions, risks, and speaker insights!

Status Python Streamlit LangGraph

🎯 What Does This App Do?

Input: A meeting transcript in plain text format
Output: Structured meeting minutes with:

  • 📋 Executive Summary - Key highlights from the meeting
  • ✅ Action Items - Tasks with owners, due dates, and priorities
  • 💡 Decisions - Important decisions made during the meeting
  • ⚠️ Risks & Concerns - Potential issues identified
  • 👥 Speaker Analysis - Who said what and their roles

📊 How It Works

Simple 6-step workflow:

Upload → Clean Text → Parse Speakers → AI Analysis → Extract Info → Display Results
  1. Upload - You provide a meeting transcript (.txt file)
  2. Clean - Remove formatting issues and normalize text
  3. Parse - Identify who said what
  4. Analyze - AI reads and understands the context
  5. Extract - Pull out actions, decisions, risks, speaker insights
  6. Display - Show results in beautiful, organized tabs

✨ Key Features

  • 📤 Simple Upload - Upload .txt transcript files with one click
  • 🤖 AI-Powered - Uses LangGraph + Ollama (TinyLlama) for intelligent analysis
  • 📋 Structured Output - Automatically extracts 5 key components
  • 🎯 Smart Detection - Finds task owners, due dates, and priorities automatically
  • 💻 Beautiful UI - Clean, modern Streamlit interface
  • ⚡ 100% Local - All processing happens on your machine (no cloud APIs!)
  • 🚀 Easy Deployment - Deploy to Streamlit Cloud in minutes
  • 📥 Export Ready - Download results as JSON

🚀 Quick Start (3 Steps!)

Step 1: Install Ollama

Ollama runs the AI model locally on your computer.

# Download and install from: https://ollama.ai/download
# Then download the TinyLlama model (small and fast)
ollama pull tinyllama

Step 2: Install Dependencies

# Navigate to project folder
cd Clarify_Meet_AI

# Install Python packages
pip install -r requirements.txt

Step 3: Run the App!

# Start Streamlit
streamlit run streamlit_app.py

That's it! Open your browser to http://localhost:8501 🎉

📖 Complete Setup Guides

📁 Project Structure

🎯 Essential Files (For Streamlit Deployment)

Clarify_Meet_AI/
│
├── streamlit_app.py              # ⭐ Main application file
├── requirements.txt              # Python dependencies
├── packages.txt                  # System dependencies
│
├── .streamlit/                   # Streamlit configuration
│   ├── config.toml              # UI theme and settings
│   └── secrets.toml             # API keys (DON'T commit!)
│
├── backend/                      # AI Processing Engine
│   ├── app/                     # Main logic
│   │   ├── langgraph_agent.py  # 🤖 Core AI agent
│   │   ├── config.py           # Settings
│   │   ├── schemas.py          # Data models
│   │   └── services/           # Utilities
│   │       ├── text_cleaner.py
│   │       ├── speaker_parser.py
│   │       └── fallback_parser.py
│   ├── models/                  # Pydantic models
│   └── utils/                   # Helpers
│
└── docs/                         # 📚 Documentation
    ├── QUICKSTART_STREAMLIT.md
    ├── STREAMLIT_DEPLOYMENT.md
    ├── DEPLOYMENT_CHECKLIST.md
    └── TESTING.md

📦 Optional Files (In extras/ folder)

Not needed for Streamlit, but available for Docker or original setup:

extras/
├── docker/                       # Docker deployment
│   ├── docker-compose.yaml
│   ├── Dockerfile
│   └── SETUP.md
├── frontend/                     # Original HTML/CSS/JS UI
├── samples/                      # Example transcripts
└── scripts/                      # Helper scripts

💡 How to Use (Step-by-Step)

Step 1: Prepare Your Transcript

Create a .txt file with speaker labels:

John: Good morning everyone. Let's start our sprint planning.

Sarah: I'll work on the login page redesign. I can finish it by Friday.

Mike: I'll handle the backend API for authentication. Due date is next Monday.

John: Decision: We will use JWT tokens for authentication.

Sarah: One risk - the design needs approval from stakeholders first.

💡 Tips for better results:

  • Use clear speaker labels (e.g., John:, Sarah:)
  • Mention dates explicitly ("by Friday", "next Monday")
  • Include "I will" statements for task owners
  • Mark decisions clearly ("Decision:", "We decided")
  • Note risks ("Risk:", "Concern:", "Issue:")

Step 2: Upload to the App

  1. Open http://localhost:8501
  2. Click "Choose a transcript file (.txt)"
  3. Select your transcript
  4. Click "🚀 Analyze Transcript"
  5. Wait 30-60 seconds for AI processing

Step 3: View Results

The app displays 6 tabs:

Tab What You Get
📋 Summary Top 3-5 key points from the meeting
✅ Action Items Tasks with owner, due date, priority, status
💡 Decisions What was decided and why
⚠️ Risks Concerns with mitigation plans
👥 Speakers Who participated and contributions
📊 Metadata Processing details and warnings

Step 4: Download Results

Click "📥 Download JSON Results" to export everything!

🎯 What Gets Extracted (Output Details)

📋 Executive Summary

  • 3-5 bullet points covering meeting highlights
  • AI-generated from full transcript context

✅ Action Items

Each action includes:

  • Description - What needs to be done
  • Owner - Who's responsible (auto-detected)
  • Due Date - Converts "tomorrow", "Friday" to actual dates
  • Priority - High/Medium/Low
  • Status - Pending (default)

Example:

{
  "description": "Work on login page redesign",
  "owner": "Sarah",
  "due_date": "2026-01-17",
  "priority": "Medium",
  "status": "Pending"
}

💡 Decisions

  • Decision - What was decided
  • Rationale - Why it was decided
  • Owner - Who made the decision
  • Context - Background information

⚠️ Risks & Concerns

  • Risk - The issue or concern
  • Impact - High/Medium/Low severity
  • Mitigation - How to address it
  • Owner - Who's responsible

👥 Speaker Spotlight

  • Name - Speaker's name
  • Role - Auto-detected (PM, Developer, QA, Designer)
  • Contribution Count - Number of times they spoke
  • Key Points - Their main contributions

🛠️ Technology Stack

Component Technology Purpose
Frontend Streamlit 1.31.0 Beautiful web interface
AI Framework LangGraph 0.0.20 Orchestrates AI workflow
LLM Ollama + TinyLlama Local AI model (no API costs!)
Language Python 3.11+ Backend processing
Deployment Streamlit Cloud Easy cloud hosting

Why These Technologies?

  • Streamlit - Makes Python apps look professional with minimal code
  • LangGraph - Manages complex AI workflows (like an assembly line)
  • Ollama - Runs AI models locally (fast, private, free)
  • TinyLlama - Small but powerful (600MB, runs on laptop)

🧠 How the AI Works (Technical Overview)

1. Text Cleaning → Remove noise, fix formatting
2. Speaker Parsing → Identify who said what
3. LLM Analysis → AI reads and understands context
4. Information Extraction → Pull out key details
5. Validation → Check for missing info, add warnings
6. Structured Output → Return organized JSON

Core Components:

  • langgraph_agent.py - Orchestrates the AI workflow
  • text_cleaner.py - Cleans up transcript text
  • speaker_parser.py - Identifies speakers
  • fallback_parser.py - Backup if AI fails

🌟 Example Use Cases

For Team Meetings

  • Sprint planning sessions
  • Retrospectives
  • Daily standups
  • Design reviews
  • Architecture discussions

For Client Meetings

  • Requirements gathering
  • Status updates
  • Stakeholder reviews
  • Decision-making sessions

For Project Management

  • Automatically track action items
  • Generate meeting summaries
  • Identify risks early
  • Monitor decision history

🐛 Common Issues & Solutions

❌ "Connection refused" or "Ollama not accessible"

Problem: Ollama isn't running
Solution:

# Check if Ollama is running
curl http://localhost:11434/api/tags

# Windows: Check system tray for Ollama icon
# Mac/Linux: ollama serve

❌ "Model 'tinyllama' not found"

Problem: Model not downloaded
Solution:

ollama pull tinyllama
ollama list  # Verify it's there

❌ "Module not found: streamlit"

Problem: Dependencies not installed
Solution:

pip install -r requirements.txt

❌ Processing takes too long

Problem: Large transcript or first run
Solutions:

  • First run is slower (model loading)
  • Keep transcripts under 5000 words
  • Close other applications

❌ Port 8501 already in use

Solution:

streamlit run streamlit_app.py --server.port 8502

📖 More help: docs/TESTING.md

🔒 Privacy & Security

✅ 100% Local Processing - Data never leaves your computer
✅ No Cloud APIs - No data sent to OpenAI, Google, etc.
✅ No Storage - Transcripts processed in-memory only
✅ Open Source - Review all code yourself

Note: For production, consider adding authentication, HTTPS, and encryption.

🎨 Customization Options

Change the AI Model

Edit backend/app/config.py:

OLLAMA_MODEL: str = "llama2"  # or mistral, codellama

Then:

ollama pull llama2

Modify UI Theme

Edit .streamlit/config.toml:

[theme]
primaryColor = "#FF4B4B"
backgroundColor = "#FFFFFF"

Adjust Extraction Logic

Edit backend/app/langgraph_agent.py to customize prompts and rules.

🚧 Roadmap & Future Plans

  • PDF/DOCX export
  • Audio transcription support
  • Multi-language support
  • Calendar integration (Google Calendar, Outlook)
  • Email notifications for action items
  • Persistent storage option
  • User authentication
  • Team collaboration features
  • Advanced analytics dashboard

🐳 Alternative: Docker Deployment

For production or API access:

cd extras/docker
docker-compose up --build

Access at: http://localhost:8000 (includes REST API)

📖 See extras/docker/SETUP.md for details

🤝 Contributing

We welcome contributions!

  1. Fork this repository
  2. Create a feature branch: git checkout -b feature/my-feature
  3. Make your changes
  4. Test thoroughly
  5. Commit: git commit -m 'Add my feature'
  6. Push: git push origin feature/my-feature
  7. Open a Pull Request

Ideas: Add features, improve prompts, enhance UI, add languages, write tests, improve docs

📝 What We Built - Complete Overview

This project transforms raw meeting transcripts into actionable insights.

Architecture

  1. Frontend - Streamlit web interface (Python-based)
  2. Backend - LangGraph agent workflow engine
  3. AI Engine - Ollama running TinyLlama locally
  4. Pipeline: Clean text → Parse speakers → AI analysis → Extract info → Validate → Output JSON

Key Innovations

  • ✅ 100% local processing (no cloud)
  • ✅ Smart date conversion ("tomorrow" → actual date)
  • ✅ Automatic owner assignment from "I'll" statements
  • ✅ Priority detection from urgency keywords
  • ✅ Role inference (PM, Developer, QA, Designer)
  • ✅ Confidence scoring
  • ✅ Fallback parser if LLM fails

Deployment Options

  1. Local - Run on laptop with Streamlit
  2. Cloud - Deploy to Streamlit Cloud (free tier)
  3. Docker - Containerized for production
  4. API - Use FastAPI backend (in extras/)

📚 Documentation

🙏 Acknowledgments

Built with amazing open-source technologies:

📞 Support & Questions

  • 📖 Documentation - Check docs/ folder
  • 🐛 Bug Reports - Open a GitHub issue
  • 💡 Feature Requests - Open a GitHub discussion
  • ❓ Questions - See docs/TESTING.md

📄 License

This project is open source and available for educational purposes.

Built with ❤️ using AI, LangGraph, and Streamlit

Status: ✅ Production-ready | Last Updated: January 2026

Repository: github.com/patilanupam/Clarify_Meet_AI

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages