Sisyphus

INSPIRATION

Small local shopkeepers manage inventory, sales, and credit manually or with complex software that doesn't fit their fast-paced lifestyle and business needs. We wanted to build a simple, conversational voice interface that acts as an intelligent talking account book. This gave rise to "Bahi". It would allow shopkeepers to record sales, track stock, handle credit, and ask business performance questions naturally in English.

FUNCTIONS?

Bahi is a talking account book for small shops powered by Model Context Protocol (MCP). It handles:

Recording: Logs cash sales, credit, stock purchases, price adjustments, and new customers.
Safety & Ledger Security: Uses confirmation safeguards where changes are read back to the user before saving. History is strictly append-only and never deleted directly.
Analytics & Restock: Answers queries like "Why were sales down on Tuesday?", identifies credit risks or slow-moving items, and generates budget-constrained restock plans.
Natural Name & Language Support: Resolves spoken names and common shop vocabulary (e.g., "cheeni", "chawal", "dal", "dedh kilo") to a limited extent.

HOW OF IT

Backend & MCP Server: Built using Node.js.
Domain Engine: TypeScript domain engine handling item inventory, double-entry ledger state, safety checks.
Simulator Frontend: Web interface simulating Alexa+ voice interactions, displaying real-time ledger updates and visual charts.

CHALLENGES

Parsing Conversational Phrasing: Normalizing natural language inputs containing multi-unit conversions (e.g., kilograms, dozens, packets) and colloquial Hinglish numbers for additional functionality.
Preventing Double-Recording: Designing state validation to ensure duplicate user confirmations ("yes", "haan") do not double-record transactions or apply expired previews.

WHAT WE ARE PROUD OF

Built a fully functional MCP server with 25 working tools operating over standard protocols.
Achieved 100% test coverage across 61 unit and integration test suites on local setups.
3.Created reliable restock planning and sales root-cause analysis engines that give actionable shop insights.

LEARNINGS

1.Designing safe tool-calling boundaries for conversational AI managing sensitive financial ledgers. 2.Implementing structured MCP protocol schemas for external client integration.

WHAT'S NEXT FOR BAHI

Native integration with real Alexa+ hardware and live OAuth authentication flows.
Stronger support for Hindi and other regional languages.
Supplier payment tracking and direct integration with digital payment platforms.