Panchayat-AI

Panchayat-AI

Inspiration

The inspiration came from a simple real-world experience:

Why should a customer have to negotiate with five different shops when an AI agent could coordinate the process?

When purchasing something from a local market, customers typically follow this process:

Find Shop A ↓ Ask Price ↓ Negotiate ↓ Remember Price ↓ Find Shop B ↓ Negotiate Again ↓ Find Shop C ↓ Compare ↓ Repeat

The problem isn't necessarily finding a shop. The problem is discovering what the market is actually willing to offer.

We wanted to build something beyond a conventional price-comparison platform.

Our idea was:

Online marketplaces compare listed prices. Panchayat-AI helps discover negotiated prices.

The name Panchayat represents bringing multiple parties together to reach an informed decision. In our system, those parties are customers, local businesses, and AI agents.

What it does

Panchayat-AI allows a customer to express a requirement naturally, for example:

"I need a laptop for coding under ₹60,000 near me."

The platform converts this request into a structured requirement and begins the discovery and negotiation workflow.

Core workflow Customer Requirement ↓ Requirement Agent ↓ Real Device Location ↓ Local Business Discovery ↓ Participating Seller Connection ↓ Seller Responses ↓ Fact Bus ↓ Negotiation Agent ↓ Offer Verification ↓ Deal Intelligence ↓ Customer Recommendation Specialized AI Agents

Requirement Agent

Converts natural-language or voice requests into structured requirements such as product, budget, location, preferences, warranty requirements, and other constraints.

Seller Discovery Agent

Uses requirement and geographic context to identify relevant local businesses.

Negotiation Agent

Uses customer constraints and verified market context to formulate negotiation actions.

Offer Extraction Agent

Converts seller responses into structured information such as:

Price Availability Warranty Terms Seller response Timestamp

Deal Intelligence Agent

Evaluates available verified offers and provides a transparent recommendation rather than simply returning the lowest number.

Fact Bus

The Fact Bus acts as the shared real-time memory of the negotiation.

For example:

Seller A → ₹62,000 Seller B → ₹61,000 Best Offer → ₹61,000 AI Counter-Offer → Sent Seller B → ₹59,500 Offer → Verified

Agents consume these structured events instead of relying only on isolated conversations.

Role-Based Platform

Panchayat-AI provides three primary roles:

Customer

Create purchase requests Use voice/text Discover nearby sellers View live offers Compare offers Track negotiations

Seller

Verify business Manage products Manage availability Receive negotiation requests Submit offers Counter offers Configure negotiation policies View analytics

Admin

Verify sellers Manage users Monitor negotiations Inspect AI decisions Review audit logs Monitor system health Access platform analytics

How we built it

We designed Panchayat-AI as a real-time, event-driven, multi-agent system rather than a simple chatbot.

Frontend

We built a modern AI-first interface using:

Next.js TypeScript Responsive UI Real-time offer updates Interactive dashboards Customer workflows Seller portal Admin dashboard

The UI focuses on making complex AI negotiations understandable to normal users.

Backend

The backend uses:

FastAPI Python REST APIs WebSockets Authentication Role-Based Access Control Event-driven services Data Layer

We use MongoDB for operational and transactional information including:

Users Seller profiles Products Negotiation sessions Seller responses Offers Verification records Fact Bus events Audit logs Location Intelligence

The platform uses real device location with user permission and business/location services to identify relevant nearby businesses.

An important architectural distinction is:

Business Discovery ≠ Seller Participation ≠ Negotiation Authorization

Finding a business through a public listing does not mean that the business has consented to automated negotiation.

Only appropriately connected and authorized sellers should participate in negotiation workflows.

AI Architecture

Our architecture separates AI reasoning from deterministic system operations:

LLM ↓ Agent Decision ↓ Tool / Service ↓ Validation ↓ Execution ↓ Verified Event

This prevents an AI-generated response from automatically becoming a trusted market fact.

Our design principle is:

LLMs propose. Deterministic services validate and execute.

Real-Time Architecture

WebSockets and event-driven updates allow negotiation states to change without requiring the customer to repeatedly refresh the application.

A negotiation can therefore progress as:

OFFER_RECEIVED ↓ OFFER_EXTRACTED ↓ OFFER_VERIFIED ↓ COUNTER_OFFER_GENERATED ↓ COUNTER_OFFER_SENT ↓ SELLER_RESPONSE ↓ BEST_OFFER_UPDATED

Challenges we ran into

Building a genuinely real-time negotiation system

A normal request-response application was insufficient because offers and seller responses can change while a negotiation is active.

We had to design real-time state synchronization using WebSockets and event-driven updates.

Preventing AI hallucinations

An AI system could potentially invent a price, availability, seller response, or warranty.

We therefore separated:

AI-generated proposals External data Seller-provided information Verified system facts

Only validated information should become part of the trusted negotiation state.

Real business discovery vs. real seller participation

One of the biggest practical challenges was understanding that discovering a local business does not automatically give us permission to negotiate with it.

This required us to design an explicit seller connection and authorization layer.

Making recommendations trustworthy

The cheapest offer is not always the best offer.

For example:

Best Deal  =Lowest Price

A recommendation may need to consider:

DealScore=f(Price,Availability,Warranty,Distance,Reliability,Preferences)

This motivated our Deal Intelligence layer.

Designing secure RBAC

We did not want role-based access to simply mean hiding buttons in the frontend.

Permissions are therefore designed to be enforced at the API and resource/service layers as well.

Building without fabricated market data

We deliberately avoided hardcoded sellers, fake prices, fake locations, and fabricated negotiation results.

The system is designed around the principle:

No fake sellers. No fake offers. No fake locations.

Accomplishments that we're proud of

🤖 Multi-Agent Architecture

We transformed a conventional price-comparison idea into a specialized AI-agent workflow involving requirement understanding, seller discovery, negotiation, offer extraction, and deal intelligence.

⚡ Real-Time Fact Bus

We developed a shared event-driven context layer that allows multiple agents to consume the same structured negotiation state.

📍 Hyperlocal Intelligence

The system combines customer requirements with geographic context to focus on relevant local businesses.

🤝 AI-Assisted Negotiation

Panchayat-AI moves beyond static price comparison toward dynamic negotiation workflows.

🔐 Role-Based Architecture

Customer, Seller, and Admin workflows are separated through proper authorization rather than UI-only restrictions.

🔎 Transparent Recommendations

Instead of simply saying "this is the cheapest", the system can explain why an available offer is competitive based on its verified attributes.

🏪 Connecting Offline Commerce with AI

Our biggest accomplishment is creating an architecture that can potentially bring fragmented local businesses into an intelligent digital negotiation network.

What we learned

AI agents are systems, not just prompts.

A powerful LLM alone cannot operate a reliable commerce platform.

Real-world agents need:

Tools APIs Databases Validation Authorization Event systems Observability Auditability Verified facts matter more than generated answers.

We learned to distinguish between:

What the AI thinks

and

What the system has actually verified.

This became one of the core architectural principles of Panchayat-AI.

Real-world AI requires strong backend engineering.

The AI component is only one part of the platform. Location, authentication, seller identity, consent, real-time communication, database consistency, security, and failure handling are equally important.

Local commerce is fundamentally different from e-commerce.

Offline markets involve:

Dynamic pricing Negotiation Relationships Local availability Trust Human communication Geographic constraints

Building for this environment required us to think beyond traditional marketplace architectures.

The best deal isn't always the cheapest price.

A slightly higher price with verified availability, warranty, reliability, and proximity can be more valuable than an unverified low price.

What's next for Panchayat-AI

Our goal is to evolve Panchayat-AI from a hackathon prototype into a local-commerce intelligence network.

🌐 1. Expand the Seller Network

Create stronger seller onboarding, verification, and authorized communication workflows.

🗣️ 2. Multilingual Voice Negotiation

Support natural voice interactions in languages commonly used in local markets, including:

Hindi Kannada Marathi Telugu Tamil Urdu English 🏪 3. Seller Integrations

Allow participating businesses to connect their own product catalogs, inventory, availability, and business systems.

📦 4. Inventory Intelligence

Move from:

"Who has the product?"

toward:

"Who has it, what can they offer, when is it available, and what are the terms?"

💳 5. Transaction Layer

Future versions could support:

Secure payments Digital invoices Order management Pickup Delivery coordination Transaction tracking 📊 6. Local Market Intelligence

With sufficient participating data and appropriate privacy protections, Panchayat-AI could provide aggregated insights into:

Demand patterns Price movements Product trends Local supply gaps Seasonal demand 🚀 Long-Term Vision

We envision Panchayat-AI becoming an AI-powered negotiation infrastructure for local commerce.

Instead of customers individually navigating fragmented markets, AI agents can help coordinate discovery, negotiation, verification, and decision-making across participating local businesses.

From fragmented local markets to an intelligent negotiation network.