Vyom — AI Business OS
Voice-First Business Operating System for Indian Retail
<150ms
Edge TTFB
95%
Hinglish Intent Match
0ms
Optimistic UI Updates
Executive Overview
Vyom is a next-generation AI Business Operating System engineered for Indian retail. Instead of forcing shopkeepers to type, Vyom lets them speak — a Web Speech API pipeline streams audio, a multilingual recognizer normalises Hindi/Marathi/Gujarati phonetics, and a custom fuzzy matcher resolves regional synonyms against live inventory. Built on TanStack Start v1 + React 19 with Supabase Postgres (RLS on every table), deployed on Cloudflare Workers.
The Problem
Traditional POS systems require manual barcode scanning or typing, slowing down checkout during peak hours.
The Solution
Voice-first billing powered by speech recognition and fuzzy matching. Automatically creates missing products and analyzes inventory.
Core Capabilities
Voice-First Billing
Real-time speech recognition in Hindi & English.
Dynamic Product Creation
AI auto-adds unknown products to inventory.
AI Business Insights
Predicts low stock and analyzes sales trends.
Smart Quantity Detection
Understands KGs, liters, dozens, and packets natively.
System Architecture
Architecture Flow
How data moves through the system, from user interaction to backend processing and response generation.
Voice Input
User taps mic → webkitSpeechRecognition boots with chosen language (en-IN / hi-IN / mr-IN / gu-IN).
Interface Gallery
Home
Technical Deep Dive
Voice Pipeline Architecture
Web Speech API captures audio → Custom VAD (Voice Activity Detection) trims silence → Transcript is tokenized → Fuzzy matcher scores against inventory → Best match is added to cart with quantity extraction — all in <200ms.
AI Billing Assistant
Co-occurrence analysis on historical bills builds a probability matrix. When 'bread' is added, the system calculates P(butter|bread) and surfaces high-probability cross-sells as smart suggestions.
Inventory Intelligence
Exponential smoothing on daily sales data predicts stock depletion dates. Dead stock detection flags items with zero movement in 30+ days. Fast-mover analysis identifies top revenue generators.