PM_Copilot

PM Copilot: A multi-agent AI assistant that generates PRDs, market sizing (TAM/SAM/SOM), user stories, and risk registers from a single product description in under 90 seconds. Built with GPT-4o, Gemini 2.0 Flash, Groq/Llama-3.3, LangChain, and Streamlit.

🚀 PM Copilot — Multi-Agent AI Assistant for Product Managers

Go from a product idea to a full PRD, market analysis, user stories, and risk register — in under 90 seconds.

[Streamlit App] (https://genai-copilot-kllgp4b2wyconulezh6fiw.streamlit.app/)

---

🔗 Live Demo

Try it now → https://genai-copilot-kllgp4b2wyconulezh6fiw.streamlit.app/

No login or setup required. Type a product idea and get four PM deliverables instantly.

---

📌 Overview

PM Copilot is a fully deployed, multi-agent AI assistant that automates the four most time-consuming Product Management deliverables:

Deliverable	Model Used	What You Get
PRD	GPT-4o (OpenAI)	Problem statement, user personas, functional & non-functional requirements, success metrics
Market Sizing	Gemini 2.0 Flash (Google)	TAM / SAM / SOM estimates, competitive landscape, go-to-market recommendations
User Stories	Groq / Llama-3.3	Role-based stories in standard format with acceptance criteria
Risk Register	Groq / Llama-3.3	Technical, market, and execution risks ranked by severity with mitigations

All four outputs are generated simultaneously from a single natural-language product description and can be refined through ongoing conversation.

---

🏗️ Architecture

┌─────────────────────────────────────────────────────┐
│                   Streamlit Chat UI                  │
│           (Input, Tabs, Session State, Export)        │
└──────────────────────┬──────────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────────┐
│              LangChain Orchestrator                  │
│     (Prompt Templates, Routing, Conversation Memory) │
└───────┬──────────────┬──────────────┬───────────────┘
        │              │              │
        ▼              ▼              ▼
   ┌─────────┐   ┌──────────┐   ┌──────────────┐
   │ GPT-4o  │   │  Gemini  │   │ Groq/Llama   │
   │  (PRD)  │   │ 2.0 Flash│   │    3.3       │
   │         │   │ (Market) │   │(Stories/Risk)│
   └─────────┘   └──────────┘   └──────────────┘

Why Three Different Models?

• GPT-4o — Best-in-class for long-context structured document generation (PRDs).
• Gemini 2.0 Flash — Fast, free-tier friendly, and strong at research-style outputs (market sizing).
• Groq / Llama-3.3 — Completely free, ultra-fast inference, and reliable at following structured templates (user stories & risk tables).

Two of the three models cost nothing, keeping the system nearly free to operate.

---

✨ Key Features

• One-Shot Generation — Describe your product idea in plain English, get four complete deliverables.
• Multi-Agent Orchestration — LangChain routes each task to the best-suited LLM in parallel.
• Conversational Refinement — Keep chatting to refine any output. No need to start over.
• Markdown Export — Download any deliverable as a .md file ready for Confluence, Notion, or GitHub.
• Live Deployment — Publicly accessible on Streamlit Community Cloud with zero setup.

---

🛠️ Tech Stack

Layer	Technology	Role
Frontend	Streamlit	Chat UI, session state, tab navigation, download buttons
Orchestration	LangChain	Task routing, prompt templating, conversation memory
LLM — PRD	GPT-4o (OpenAI)	Structured PRD generation, multi-turn dialogue
LLM — Market	Gemini 2.0 Flash (Google)	TAM/SAM/SOM, competitive analysis, GTM strategy
LLM — Stories/Risk	Groq / Llama-3.3	User stories, risk register, structured tables
Infrastructure	Streamlit Cloud + GitHub	Public deployment, version control, secret management

---

⚡ Quick Start

Prerequisites

• Python 3.10+
• API keys for OpenAI, Google Gemini, and Groq

Environment Variables

Create a .env file in the project root with the following:

OPENAI_API_KEY=your_openai_key_here
GOOGLE_API_KEY=your_gemini_key_here
GROQ_API_KEY=your_groq_key_here

⚠️ Never commit your .env file. The .gitignore is already configured to exclude it.

---

📊 Sample Output

For the prompt: "A productivity app for college students with AI study schedules"

Deliverable	Highlights
PRD	4 user personas, 12 functional requirements, 5 success metrics
Market Analysis	TAM: $8.4B · SAM: $1.2B · SOM: $84M · 4 competitors identified
User Stories	10 role-based stories across student, professor, and admin roles
Risk Register	7 risks ranked High/Medium/Low with mitigation strategies

All generated in under 90 seconds.

---

🧩 Challenges & Learnings

• Gemini quota limits — Free tier ran out mid-demo; added graceful fallback messaging.
• Invisible characters in API keys — Hidden newline characters caused silent auth failures; solved with .strip() on all key loads.
• GitHub secret scanning — Hardcoded keys in notebooks triggered push blocks; migrated all secrets to Streamlit's secrets manager.
• LangChain import paths — Always import from langchain_core.messages, not the deprecated langchain.schema.

---

🔮 Future Roadmap

• Real-time web search integration for live market data
• User authentication and saved project history
• Direct export to Jira, Notion, and Confluence
• Fine-tuned models trained on PRD-specific corpora
• Competitive battle cards and OKR generation modules

---

📄 License

This project is licensed under the MIT License.

---

⭐ If you found this useful, consider giving the repo a star!