🚀 Deck Insight

Deck Insight is an open-source AI tool to analyze, summarize, and score startup pitch decks automatically.

• 📄 Upload a startup's pitch deck (PDF)
• 📋 Extracts and structures key information
• 🏆 Scores the startup based on a custom evaluation rubric (for investors, accelerators, competition judges)
• 🌎 Calculates market size estimates from real-time internet data
• 🛠️ Pulls insights from the startup’s GitHub repositories (for developer tools/open-source startups)
• 💬 Includes a QA chatbot to ask natural language questions about the startup

Fully extensible and customizable for your workflows.
Perfect for investors, analysts, startup competitions, accelerators, and ecosystem builders.

Observability on AI agents are implemented through LangSmith.

---

🛠 Output Preview

Note: This repository contains only the backend implementation. The screenshots below are from a separate frontend integration built using Lovable, showcasing how the API endpoints can be consumed and displayed in a user interface.

---

🛠 How It Works

1. Upload or provide a link to the startup’s pitch deck (PDF).
2. The system extracts key information and fills a structured profile.
3. Applies a scoring rubric customized to your preferences.
4. Estimates market size from online data sources.
5. (Optional) If it's a developer tool, pulls additional GitHub repo data.
6. Ask the built-in chatbot for any questions about the startup profile.

---

📦 Installation

1. Clone the repository:

git clone https://github.com/hitesh-ag1/deck-insight.git
cd deck-insight

2. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate

3. Install the dependencies

pip install -r requirements.txt

4. Set up the environment variables:

cp .env.example .env

Add your API keys to .env

5. Run FastAPI server:

uvicorn main:app --reload

6. Test the API with sample decks under examples folder:

# Use any deck from the examples/ directory to test
curl -X POST "http://localhost:8000/analyze-complete" \
  -H "Content-Type: application/json" \
  -d '{"file_path": "examples/MapMe-Pitch-Deck.pdf"}'

# Or use the Swagger UI at http://localhost:8000/docs to test interactively

---

🏗️ How It’s Built

Startup Deck Copilot is architected using modular AI agents orchestrated with the LangGraph framework, all packaged behind a FastAPI server for easy deployment and API access.

✨ Core Components

• Pitch Deck Scorer

  • Inputs images of pitch deck slides
  • Uses Gemini 2.0 Flash for OCR and visual understanding
  • Generates a structured summary of key startup information (problem, solution, traction, market, business model, team, funding) using GPT-4o-mini
  • Scores the startup against a predefined, customizable rubric

• Market Research Agent

  • Uses Tavily Search to research sector, market size, and competitor landscape
  • Summarizes search results using GPT-4o-mini

• GitHub Viewer Agent

  • Uses Firecrawl to scrape the startup’s GitHub repositories
  • Summarizes repository activity, health, and community engagement

• QA Chatbot

  • Built as a Retrieval-Augmented Generation (RAG) system
  • Indexes startup profiles into Elasticsearch
  • On query, retrieves most relevant context and answers using GPT-4o-mini

• Supervisor Agent

  • Orchestrates the full analysis pipeline:
    1. Pitch Deck Scorer
    2. Market Research Agent
    3. GitHub Viewer Agent (conditional)

---

🛠️ Stack

Layer	Technology
Vision Model	Gemini 2.0 Flash
Language Model	GPT-4o-mini
Agent Orchestration	LangGraph
API Server	FastAPI
Web Search	Tavily API
Web Scraping	Firecrawl
Vector Database (RAG)	Elasticsearch

---

📚 API Endpoints

• POST /analyze-complete: Full pitch deck analysis
• POST /analyze-pitch-deck: Pitch deck scoring and summary
• POST /analyze-market-size: Market research analysis
• POST /analyze-github-repository: GitHub repository evaluation
• POST /chat-assistant: Interactive Q&A about the pitch deck

---

📂 Project Structure

pitch_deck/
├── .env.example         # Example environment variables template
├── README.md             # Project documentation
├── main.py               # FastAPI application entry point
├── requirements.txt      # Project dependencies
├── core/                 # Core functionality and utilities
│   ├── prompts.py        # AI model prompts and templates
│   ├── schema.py         # Shared data models and schemas
│   ├── settings.py       # Application configuration
│   └── utils.py          # Shared utility functions
└── agents/               # AI agents for different analysis tasks
    ├── <agent_name>/     # Each agent has a modular folder
    │   ├── agent.py      # Agent logic and workflow
    │   ├── helpers.py    # Helper functions (optional)
    │   ├── models.py     # Agent-specific data models
    │   └── nodes.py      # LangGraph nodes defining steps

---

🧩 Future Roadmap

• Advanced financial modeling (projections, valuation sanity checks)
• Comprehensive prompt testing using Promptfoo
• API and plugin integrations (e.g., Crunchbase, LinkedIn)
• Chrome extension for sourcing decks from web
• Support for multiple decks comparison

---

Built with ❤️ to make startup evaluation faster, fairer, and more data-driven.