🧠 BizRadar AI

AI-Powered Startup Intelligence & Business Analysis Agent

Transform any startup idea into a structured business intelligence report — powered by a hand-built ReAct agent, parallel tool execution, real-time web search, and RAG-grounded document analysis.

Getting Started • Features • Architecture • Roadmap • Contact

---

🎯 The Problem This Solves

You have a startup idea. You need answers — fast, cited, and trustworthy.

The problem with generic AI tools: they hallucinate. They produce confident market statistics that don't exist. They cite reports that were never written. You can't put that in an investor pitch deck.

BizRadar AI solves this with three layers of grounding:

• Real-time web search via Tavily — every market claim has a source URL
• Document-grounded RAG via ChromaDB — answers from your own pitch deck, not the LLM's memory
• ReAct reasoning loop — the agent reasons about what it needs before acting, not after

Example: "AI-powered tiffin service for students"

Output: A complete startup analysis — market potential, competitors, MVP features, tech stack, risk assessment — every claim cited, every document answer grounded.

✨ Features

📁 Project Structure

bizradar-ai/
│
├── 🤖 agent.py                # ReAct agent — Groq LLM + parallel tool execution loop
├── 🖥️  app.py                  # CLI entry point + PDF ingestion trigger
├── 🧠 context_manager.py      # Conversation memory — last 6 turns sliding window
├── 🛠️  tools.py               # Tool layer — Tavily search + Gemini analysis (self-summarizing) + RAG search
├── 📋 tools_description.py    # Tool schemas for LLM tool-calling (JSON format)
├── 📝 prompts.py              # System prompt + output format templates
├── 🗄️  rag.py                  # RAG pipeline — ingest, embed, store, query (Stage 4)
│
├── 📁 database/
│   └── chroma_db/             # Persistent ChromaDB vector store (local, gitignored)
│
├── 📦 requirements.txt        # Python dependencies
├── 🔒 .env                    # API keys (never committed)
├── 🚫 .gitignore              # Ignores .env, database, and local files
├── 📖 README.md               # This file
├── 🧭 ROADMAP.md              # Phase-by-phase build and learning path
├── 🏗️  ARCHITECTURE.md         # Deep dive into every design decision
├── 📓 LEARNING_LOG.md         # Personal learning tracker and mistake log
└── 📋 CHANGELOG.md            # Version history

🔁 ReAct Pattern — How The Agent Thinks

🏗️ System Architecture

┌─────────────────────────────────────────────┐
│                   USER INPUT                │
│         "AI tiffin service for students"    │
└──────────────────────┬──────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────┐
│           CLI INTERFACE  (app.py)           │
│     PDF ingestion trigger at startup        │
└──────────────────────┬──────────────────────┘
                       │
                       ▼
┌──────────────────────────────────────────────────────┐
│              ReAct AGENT LOOP  (agent.py)            │
│                                                      │
│  ┌──────────────┐     ┌────────────────────────────┐ │
│  │   Context    │     │       Groq LLM             │ │
│  │   Manager   │────▶│    Llama 3.3 70B           │ │
│  │  Last 6 msgs │     │  Reason → Act → Observe   │ │
│  └──────────────┘     └────────────┬───────────────┘ │
│                                    │                  │
│                    ┌───────────────▼─────────────┐   │
│                    │    ThreadPoolExecutor        │   │
│                    │    (Parallel Execution)      │   │
│                    │                             │   │
│                    │  analyze_market()            │   │
│                    │  search_knowledge_base()     │   │
│                    │  suggest_mvp()               │   │
│                    │  recommend_tech_stack()      │   │
│                    │  risk_analysis()             │   │
│                    │  search_documents() ← RAG    │   │
│                    │                             │   │
│                    │  Tavily · Gemini · ChromaDB  │   │
│                    └───────────────┬─────────────┘   │
└────────────────────────────────────┼─────────────────┘
                                     │
               ┌─────────────────────┴──────────────────┐
               │                                        │
               ▼                                        ▼
┌──────────────────────────┐           ┌───────────────────────────┐
│   STRUCTURED REPORT      │           │   RAG PIPELINE (rag.py)   │
│                          │           │                           │
│  # Market Potential      │           │  ingest_pdf()             │
│  # Competitors           │           │  embed_and_store()        │
│  # Suggested MVP         │           │  query_rag()              │
│  # Tech Stack            │           │                           │
│  # Risks                 │           │  ChromaDB Persistent      │
│  # Cited Sources         │           │  gemini-embedding-001     │
└──────────────────────────┘           └───────────────────────────┘

⚙️ Tech Stack

🚀 Getting Started

Prerequisites

• Python 3.10+
• Groq API Key — free tier available
• Tavily API Key — free tier available
• Gemini API Key — free tier available

1. Clone the Repository

git clone https://github.com/ankush-poonia007/STARTUP-AI-AGENT.git
cd STARTUP-AI-AGENT

2. Install Dependencies

pip install -r requirements.txt

3. Configure Environment Variables

# Create .env file in project root
GROQ_API_KEY=your_groq_key_here
TAVILY_API_KEY=your_tavily_key_here
GEMINI_API_KEY=your_gemini_key_here

⚠️ Never commit your .env file. It is gitignored by default.

4. Run BizRadar AI

python app.py

💬 Example Session

🚀 BizRadar AI Started
Type 'exit' to quit.

Do you have a document to upload? (YES / NO): yes
Enter Your File Path: ./pitch_deck.pdf
✅ Data ingestion complete. Data saved successfully.

You: What is the revenue projection in my pitch deck?

⚙️  Stage 1 of 3 — Executing tools...
   🔧 analyze_market()
   🔧 search_knowledge_base()
   ✅ Stage 1 complete.

⚙️  Stage 2 of 3 — Executing tools...
   🔧 suggest_mvp()
   🔧 recommend_tech_stack()
   ✅ Stage 2 complete.

⚙️  Stage 3 of 3 — Executing tools...
   🔧 risk_analysis()
   ✅ Stage 3 complete.

🔍 Stage 4 — Querying your document...
   🔧 search_documents()
   ✅ Report ready.


📊 BizRadar AI:

## Market Potential
High demand in tier-1 college cities. [Source: economictimes.com]

## Suggested MVP
- User registration & meal preferences
- Daily menu with AI personalization
- Subscription billing

## Recommended Tech Stack
- Backend: FastAPI  |  Frontend: React
- Database: PostgreSQL  |  AI Layer: Gemini API

## Risks
- High CAC in student market
- Competition from Swiggy/Zomato

## Final Verdict
Viable niche opportunity with strong MVP potential.

🧭 Roadmap

🏛️ Design Philosophy

BizRadar is intentionally built without LangChain or LlamaIndex to deeply understand how AI agents work at the foundational level. Every component — the ReAct loop, tool calling, parallel execution, RAG pipeline, vector store — is implemented manually.

When you eventually use a framework, you will understand exactly what it is abstracting and why.

🎯 Learning Objectives

• ✅ Prompt Engineering & System Design
• ✅ Context Window Management
• ✅ Tool-Augmented AI Agents
• ✅ ReAct Agent Pattern
• ✅ Parallel Tool Execution
• ✅ OOP Architecture for AI Systems
• ✅ Multi-Provider LLM Integration
• ✅ Vector Embeddings & Semantic Search
• ✅ RAG Pipeline from Scratch
• ✅ ChromaDB & Persistent Vector Storage
• 🔜 Multi-Document RAG & Evaluation
• 📋 Multi-Agent Orchestration
• 📋 Production AI Engineering

📜 License

MIT License — free to use, modify, and build on.