Back to Cluster
Project Specification v1.0

VyapaarSetu: AI Commerce Agent

VyapaarSetu: AI Commerce Agent
01

The Bottleneck

"Millions of small-scale sellers in India are excluded from the digital economy due to technical complexity and language barriers. The process of listing products, managing inventory, and communicating with customers online is often overwhelming for non-tech-savvy individuals."

02

Key Features

AI Vision: Automated attribute extraction from product photos

Voice-First Interface: Native support for 9+ Indic languages

Autonomous Cataloging Agent with Dynamic LLM Pricing

Event-Driven Microservices orchestrated via Apache Kafka

WhatsApp Commerce Integration for Seamless Buyer Connection

Offline-First Capability for Low-Bandwidth Environments

03

The Architecture

I architected an autonomous agent system that uses Vision AI for 'Photo-to-Listing' automation and a multilingual Voice AI interface. Built on an event-driven microservices architecture with Apache Kafka, the system orchestrates multiple AI agents to handle cataloging, pricing, and customer support. The mobile-first interface supports 9+ regional languages, allowing sellers to manage their entire store via voice commands.

04

Overcoming Challenges

Linguistic Nuance: Tuning STT models (Whisper/IndicWav2Vec) to handle regional dialects with over 90% accuracy.

Distributed State: Managing consistent catalog states across multiple autonomous agents in a decentralized system.

Mobile Performance: Optimizing heavy AI-driven features to run smoothly on mid-range Android devices.

05

Protocol & Process

Phase 01

Phase 01: Field research to understand the listing bottlenecks for local sellers.

Phase 02

Phase 02: Designing the multi-agent orchestration layer and event schema.

Phase 03

Phase 03: Developing the Vision AI pipeline for automated attribute extraction.

Phase 04

Phase 04: Integrating the multilingual voice engine and WhatsApp gateway.

Phase 05

Phase 05: Pilot testing with local artisans to refine the voice interaction flow.

Engineered Impact

Creating a pathway for small businesses to enter the ONDC ecosystem with zero technical friction. The platform has the potential to multiply a local seller's customer base by providing them with the same technological leverage as large-scale enterprise retailers.

Tech Stack

PythonKafkaPostgreSQLReact NativeVision AILLMs
Verified Production Deployment