RELAY //

Give AI a high-level mission. It plans, discovers suppliers, makes outbound calls, negotiates prices, and locks in the deal.

Comment

RELAY // Autonomous Business Operations Agent

Inspiration

Most voice AI applications simply attach a speech interface to an LLM so a human can talk to an AI assistant. RELAY flips this paradigm: AI makes outbound phone calls on behalf of the business to execute real-world operations.

When a company needs 500 chairs delivered by Friday under $15,000, standard web APIs fail because over 80% of wholesale suppliers and local logistics operators only conduct business over the phone. RELAY bridges this execution gap using CALL-E.

What It Does

RELAY is an autonomous multi-call operational engine that runs multi-stage mission loops:

  1. Natural Language Mission Planning: Parses goals into strict operational constraints (budget ceilings, delivery SLAs, locations, and approval thresholds).
  2. Autonomous Multi-Call Telephony via CALL-E: Discovers candidate vendors and places structured voice calls to extract pricing, lead times, warranties, and stock levels.
  3. Decision Engine & Algorithmic Scoring: Evaluates all gathered quotes against mission constraints.
  4. Targeted Leverage Negotiation: Detects price discrepancies between competing quotes and automatically places a targeted second call to negotiate volume discounts.
  5. Human-in-the-Loop Safety Gate: Enforces financial approval policies before executing binding purchase commitments.
  6. Binding Confirmation Lock: Dispatches the final call to lock in purchase orders, delivery commitments, or appointment slots.

Supported Workflows

  1. Commercial Procurement (Multi-vendor inquiry, leverage negotiation, PO confirmation)
  2. Emergency Logistics Rescue (Sequential dispatcher dialing, 2-hour arrival SLA lock)
  3. Turnkey Equipment Bidding (Comparative 50kVA generator bidding, specification normalization)
  4. Priority Waitlist Scheduling (Sequential appointment backfilling)

How We Built It

  • Backend: FastAPI, SQLAlchemy, Async EventBus for real-time WebSocket streaming, and CALL-E Python SDK v0.2.
  • Frontend: Next.js 15 App Router, Tailwind CSS, custom CRT matrix aesthetics, and dynamic audio waveforms.
  • Hosting: Backend deployed on Render with automated self-ping keep-alive; Frontend deployed on Vercel with Web Analytics and Speed Insights.

What's Next for RELAY

  • Multi-party conference escalation calls.
  • Automated ERP and SAP webhook sync upon Purchase Order generation.
  • Expanded multi-language negotiation scripts for global freight and trade.

Built With

Share this project:

Updates

Submission history