Uddyog-CoPilot

💡 Inspiration

Starting a business in Bangladesh is hard, but validating a startup idea with limited mentorship and guidance is even harder. Most early-stage Bangla-speaking entrepreneurs struggle to transform vague concepts into structured, risk-tested execution plans. We built Uddyog-CoPilot to democratize startup consulting—giving founders an AI-powered strategic co-pilot that understands their local market, operational realities, and native language.

🚀 What it does

Uddyog-CoPilot is a full-suite digital business assistant tailored for Bangla-speaking entrepreneurs. Key features include:

• 🛡️ War Room (Multi-Agent AI Debate): Simulates a boardroom discussion with 3 distinct AI agent personas—a Skeptic (risk validator), a Strategist (growth advisor), and an Operator (logistics planner).
• 📍 Assumption Map & Risk Heatmap: Automatically categorizes business claims into VALIDATED, UNVALIDATED, and NEEDS_INFO for visual risk tracking.
• 📄 Document Advisor (RAG & GCS): Enables founders to upload pitch decks or business PDFs directly to Google Cloud Storage (GCS) and query an AI advisor trained on project documents.
• 💳 x402 Micropayments & Ad Monetization: Integrates Web3 micro-transactions to unlock high-tier AI analysis on demand alongside Google AdSense test units.

🛠️ How we built it

• Frontend: Next.js 15 (App Router), React 19, TypeScript, and Tailwind CSS 4. Used @xyflow/react (React Flow) for interactive assumption maps.
• Backend: Express 5 and TypeScript managed with the PM2 process manager (ecosystem.config.cjs).
• AI Engine: Powered by Google Gemini 2.5 Flash for low-latency multi-agent reasoning and localized business logic.
• Cloud Infrastructure (GCP): Integrated Google Cloud Storage (GCS) for secure document ingestion, handling pitch decks and business assets used by the RAG system.
• Database & Vector Search: PostgreSQL orchestrated via Prisma 6 with the pgvector extension for localized document embeddings.
• Monetization: Built an HTTP 402 Payment Required middleware using the x402 Micropayment Protocol.

🧠 Challenges we ran into

• Persona Consistency in Native Language: Ensuring Skeptic, Strategist, and Operator agents maintained strict persona logic during complex debates without hallucinating in native Bangla.
• x402 Payment Middleware Integration: Intercepting premium AI requests on Express 5 to trigger an automated payment challenge and verify micro-transactions before serving API responses.
• RAG & Cloud Integration: Structuring smooth document ingestion pipelines connecting Next.js, GCS buckets, and pgvector embeddings under strict time limits.

🏆 Accomplishments that we're proud of

• Successfully engineered a working Multi-Agent AI Debate System that provides multi-perspective startup analysis in native Bangla.
• Integrated Google Cloud Storage (GCS) to ensure enterprise-grade document handling for RAG workflows.
• Built a production-ready, full-stack architecture combining Next.js 15, Prisma 6, pgvector, and Gemini APIs despite severe resource constraints.

📖 What we learned

• Deepened our understanding of Multi-Agent orchestration and prompt structuring for decision-making workflows.
• Mastered RAG implementation using GCS and pgvector for localized document retrieval.
• Explored real-world micro-transactions (x402 protocol) for monetizing AI endpoints.

🔮 What's next for Uddyog-CoPilot

• 🗣️ Voice Command Navigation: Adding native Bangla voice-to-text support for regional entrepreneurs.
• 📊 Automated Financial Forecaster: Enabling founders to generate 3-year cash flow projections using AI-driven market data.
• 📱 Mobile App (PWA): Bringing Uddyog-CoPilot to mobile devices for seamless on-the-go startup management.

Built With

• apis
• cloud-services
• databases
• express.js
• frameworks
• gcp
• gemini-api
• google-cloud
• next.js
• pgvector
• platforms
• typescript