Udyara: Cultivating Women Led Enterprises

Trustworthy RAG-Based Policy Navigation System for Women-Led Startups

Udyara is a policy-aware intelligent system designed to help women entrepreneurs in India understand, verify, and navigate government startup incentives — starting with the Stand-Up India Scheme.

Built to demonstrate:

• Retrieval-Augmented Generation (RAG)
• Trustworthy, source-backed responses
• Semantic search over policy documents
• Explainable decision support

---

🚩 Problem Statement

• Government startup policies are complex, fragmented, and difficult to interpret.
• Women entrepreneurs often miss out on benefits.
• This gap results in underutilization of public funds, delayed entrepreneurship, and reduced participation of women in the startup ecosystem.

---

💡 Solution

Udyara acts as a Policy Navigator System that:

• Retrieves official policy documents
• Understands eligibility criteria
• Guides users through benefits & next steps
• Provides transparent, explainable responses

---

🧠 Core Features

• 📄 RAG-based policy understanding system
• 🔍 Eligibility reasoning
• 🧾 Source-backed answers (no hallucination)
• 🔐 Trust-first architecture

---

🏗 Architecture Overview

User Query
   ↓
Frontend (React / Next.js)
   ↓
FastAPI Backend
   ↓
RAG Pipeline
   ├── Policy PDFs
   ├── Sentence-Transformer Embeddings
   ├── FAISS Vector Store
   └── Gemini LLM (response synthesis)

• LLMs are used for interpretation and explanation, not decision-making
• The system focuses on retrieval and explanation, not autonomous decision-making or action execution.

🔀 Flow

PDF (policy)
   ↓
Local Embeddings (sentence-transformers)
   ↓
FAISS Vector Store (files on disk)
   ↓
Retriever
   ↓
Gemini LLM (answers)

---

🧪 Demo Use Cases

These queries demonstrate real-world policy discovery, eligibility validation, and benefit awareness for first-time women founders.

• "Am I eligible for Stand-Up India as a woman founder?"
• "What documents are required?"
• "What loan benefits are provided?"
• "What are common rejection reasons?"

---

🚀 Tech Stack

Layer	Technology Used	Purpose
Frontend	React	User interaction (chat-based UI)
Backend	FastAPI	API handling & orchestration
AI Pipeline	LangChain	RAG workflow management
Embeddings	Sentence Transformers	Policy document vectorization
Vector DB	FAISS	Fast similarity search
LLM	Gemini 2.5	Answer synthesis & formatting
Data Source	Govt Policy PDFs	Trusted knowledge base

---

📦 Project Structure

Udyara/
├── frontend/                 # React UI - Chat interface
│   ├── src/
│   │   ├── components/       # Navbar, Footer components
│   │   ├── pages/           # Home, Agent, About, Features pages
│   │   └── services/        # API calls to backend
│   └── package.json         # Frontend dependencies
│
├── backend/                  # FastAPI application
│   ├── app/
│   │   ├── agent.py         # Core agent logic (question answering)
│   │   ├── config.py        # Configuration (API keys, settings)
│   │   ├── main.py          # FastAPI routes and endpoints
│   │   ├── data/            # Raw PDF files for ingestion
│   │   ├── rag/             # RAG pipeline
│   │   │   ├── ingest.py    # PDF processing and chunking
│   │   │   ├── retriever.py # Semantic search retrieval logic
│   │   │   └── qa_chain.py  # Question-answering chain
│   │   └── vectorstore/     # FAISS vector database
│   │       └── index.faiss  # Persisted embeddings
│   └── requirements.txt     # Backend dependencies
└── README.md

---

⚙️ Running Locally

Option 1 — Docker (Recommended, Single Command)

Prerequisites: Docker Desktop installed and running.

# 1. Clone the repo
git clone https://github.com/your-username/udyara.git
cd udyara

# 2. Create your .env file (never commit this!)
cp .env.example .env
# Edit .env and add your real GEMINI_API_KEY

# 3. Start everything — backend + frontend in one command
docker compose up --build

• 🌐 Frontend → http://localhost:5173
• ⚙️ Backend API → http://localhost:8000
• 📖 Swagger UI → http://localhost:8000/docs

First build takes ~5 minutes (downloads model + dependencies). Subsequent runs use cached layers and start in seconds.

---

Option 2 — Manual (without Docker)

Click to expand manual setup

Backend

cd backend
python -m venv .venv
.venv\Scripts\activate          # Windows
# source .venv/bin/activate     # macOS/Linux
pip install -r requirements.txt
# Create .env with GEMINI_API_KEY=your_key
uvicorn app.main:app --reload
# Running at http://127.0.0.1:8000

Frontend (in a new terminal)

cd frontend
npm install
npm run dev
# Running at http://localhost:5173

---

🚀 Cloud Deployment (Free)

Udyara deploys for free using Render (backend) + Vercel (frontend).

Browser → Vercel (React) → Render (FastAPI + RAG)

Backend → Render

1. Push this repo to GitHub
2. Go to render.com → New → Web Service
3. Connect your GitHub repo
4. Configure:
   • Root Directory: (leave blank — uses render.yaml)
   • Runtime: Docker
   • Dockerfile Path: ./Dockerfile.backend
5. Add environment variable:
   • GEMINI_API_KEY = your Gemini API key
6. Click Deploy → copy the URL (e.g. https://udyara-backend.onrender.com)

⚠️ Free tier spins down after 15 min of inactivity. First request after sleep takes ~30s.

Frontend → Vercel

1. Go to vercel.com → New Project → import GitHub repo
2. Set Root Directory to frontend
3. Add environment variable:
   • VITE_API_URL = https://udyara-backend.onrender.com (your Render URL)
4. Click Deploy → your app is live! 🎉

---

🌱 Future Scope

• Add more women-centric policies (MUDRA, PMEGP, Startup India)
• Multilingual support (Hindi / regional languages)
• Voice-based interaction for accessibility
• State-specific incentive mapping
• Verifiable credentials for document checks

---

✨ Why Udyara Stands Out

• Eliminates policy confusion using semantic search
• Ensures zero hallucination with source-backed responses
• Designed specifically for women entrepreneurs in India
• Bridges the gap between policy availability and accessibility