Vakil: Bounded AI-to-AI Negotiation | Razorpay Buildathon 2026

Built for the Razorpay Buildathon 2026 (Track 01: AI Growth & Agentic Commerce), Vakil is a proof-of-concept platform demonstrating autonomous agent-to-agent transactions with provable constraints and complete auditability. Core Architecture & Capabilities Propose → Verify → Execute: Separates decision-making (LLM agents) from enforcement (deterministic code gates). Buyer Vakil: Autonomous purchasing agent that negotiates strictly within delegated budget mandates (max total spend, max unit price, target quantity). Merchant Vakil: Autonomous sales agent negotiating within designated pricing corridors (list price ceiling and floor price minimums) with volume bundle support. Deterministic Policy Gates: Hard constraint enforcement written in code rather than prompt instructions to prevent halluncinated deals or policy breaches. Immutable Audit Ledger: Every turn, decision rationale, and gate check is logged in an append-only ledger for full post-deal transparency. Settlement Integration: Real-time Razorpay order creation triggered automatically upon deal convergence, confirmed using signed webhooks. Technology Stack Frontend: React, Vite, Tailwind CSS v4 Backend: Node.js, Express, TypeScript Database: PostgreSQL (Neon) AI Engine: Groq (gpt-oss-120b) Payments: Razorpay API & Webhooks #RazorpayBuildathon #AgenticCommerce #AIAgents #ArtificialIntelligence #WebDevelopment #TypeScript #Groq #Razorpay