Kural – hands-free voice copilot for foundry operators
AI for foundries: catch off-spec chemistry before the melt, plan corrections, predict properties, with a hands-free voice add-on.
Built at AgentX Hackathon 2026
Team Autom Squad: Manish Raghul S J, Nethra Venkat
Built with React, Vite, Python, FastAPI, Node.js, Express, Machine Learning, Reinforcement Learning, OPC UA, Gnani Prisma STT, Gnani Timbre TTS, Vitest, Pytest, Playwright
About the project
Problem: In an induction foundry, every heat is a costly decision. Scrap chemistry varies, spectrometer readings arrive fast, and a wrong charge or a missed off-spec element can mean a rejected casting, re-melting energy and lost margin. The data exists across instruments, stock and quality records, but decisions are still made by hand.
Solution: MetalliSense is an end-to-end AI copilot for foundry melt operations, from scrap yard to sales:
- Pre-production: yard inspection with handheld XRF/LIBS, supplier scorecards, raw-material stock and inventory forecasting.
- Melt: live spectrometer readings over OPC, grade specifications, and a pre-melt gate that marks each heat GO, WARN or HOLD before charging.
- AI analysis: composition deviation (anomaly) detection, a melt correction planner, an adaptive strategy optimizer for additions, full analysis and an AI agent that explains its reasoning.
- Post-production: mechanical-property prediction, energy per heat, cost analysis and sales/dispatch.
Add-on: Kural, a hands-free voice layer (Tamil, Hindi, English) built on Gnani Prisma and Timbre, lets operators use the same pages and actions by voice, with spoken confirmations and safety gates that voice can never bypass.
Impact: fewer off-spec heats, faster corrections, and decisions grounded in live plant data, with audit trails for every action. Verified with 2,600+ automated tests across the app.