ProcureVoice

ProcureVoice turns a spoken sentence into an approval-ready procurement request. An employee says: "I need three monitors at ₦100,000 each from Techmart, approved vendor, normal request." ProcureVoice listens, detects what's missing, asks follow-up questions, checks company procurement policy, calculates the total spend, and creates a structured purchase request with a unique ID — all in a natural voice conversation, without a form. Built on the AssemblyAI Voice Agent API, ProcureVoice runs a single WebSocket carrying real-time speech recognition, LLM reasoning, text-to-speech, turn detection, and JSON-schema tool calling. Two backend tools — check_purchase_policy and create_purchase_request — execute against a Node.js server that evaluates spend thresholds, vendor status, category rules, and urgency. The result is a request that arrives ready for human approval, routed to the right level: manager, finance, or executive. The agent never approves purchases. It prepares and routes them. That boundary is deliberate: it makes ProcureVoice deployable in enterprises where autonomous spend decisions are unacceptable. Voice is the natural interface for procurement because procurement begins as conversation. Forcing employees to translate that conversation into a form is the source of most friction in the process. ProcureVoice removes the translation step. Stack: AssemblyAI Voice Agent API, WebSocket, Node.js, Express 5, Web Audio API, AudioWorklet, vanilla JS. No frameworks, no bundler, no database — deployed on Render, built entirely from a phone in Termux.