Lead Rescue Voice

An Alexa+ style agent that finds leads going cold and turns them into actionable follow-ups.

Comment

Why it exists

Small service businesses lose good opportunities when the owner is in the field instead of watching an inbox or CRM. Lead Rescue Voice helps recover those moments without giving an AI authority it should not have.

What it does

Lead Rescue Voice is an Alexa+-track lead-recovery agent that:

ranks leads by rescue urgency with transparent reasons; lets a user inspect the evidence behind each priority; prepares a safe next follow-up; treats urgent service requests as time-sensitive without inventing availability; detects price-match requests and escalates them instead of inventing discounts or changing pricing authority.

The Alexa+ style browser simulation lets a judge run the morning brief, inspect an urgent lead, trigger the price-match safety boundary, and view the MCP-style tool trace.

Technical implementation

The Amazon-specific implementation is a self-hosted Model Context Protocol server built with the current MCP Python SDK v2 and Streamable HTTP.

MCP tools:

prioritize_leads inspect_lead prepare_follow_up

MCP prompt:

morning_rescue_brief

The real HTTP smoke test connects to /mcp, calls a tool, and asserts the negotiated MCP protocol revision is 2025-11-25 or later.

The branch also includes deterministic unit tests, adversarial tests, dependency auditing, Bandit, CodeQL, and a reusable release gate. A security review caught an unsafe default all-interface bind; the server now defaults to 127.0.0.1 and requires an explicit HOST value for public deployment. The functional and security suites pass after the fix.

The rescue score is also safety-first: a critical service need gets a 90-point floor so a larger but noncritical opportunity cannot outrank an urgent customer-safety problem. A regression test verifies the browser demo scores stay synchronized with the MCP core.

Product design

The interaction is intentionally simple: show the rescue queue, explain why a lead matters now, prepare the next safe response, and stop when authority belongs to the owner.

Synthetic data keeps the demo reproducible. The agent does not autonomously send messages or invent appointment slots, discounts, warranty exceptions, or unusual commitments.

Potential impact

The same bounded rescue workflow can be adapted to HVAC, roofing, electrical, landscaping, garage-door, junk-removal, and other field-service businesses where missed follow-up directly costs revenue.

A separate public commercial proof-of-concept for after-hours HVAC lead routing is already live at: https://ava-hvac-lead-router.vercel.app

That site is not presented as the Amazon MCP demo; it is a separate proof that this customer problem can be packaged as a standalone workflow.

The commercial fork is a subscription or fixed-setup product that helps owners recover valuable inquiries without turning pricing authority over to an agent.

Why the idea is different

The differentiator is not “AI follow-up” alone. Lead Rescue Voice combines transparent rescue scoring with an explicit authority boundary: routine work can move forward, but owner-only decisions fail closed and are escalated.

Significant update disclosure

This repository existed before the hackathon. To avoid changing the project already under judging elsewhere, the Amazon work is isolated on the public amazon-alexa branch.

The branch adds the Alexa+/MCP server, Streamable HTTP implementation, protocol compatibility test, browser simulation, adversarial and security testing, friction log, demo plan, and submission-quality tooling during the Amazon Developer Hackathon window.

Open Source mini challenge

The public amazon-alexa branch is the hackathon-window open-source contribution. It includes source, tests, documentation, and the repository's MIT license.

Contribution: https://github.com/jpadilla987-png/lead-rescue-agent/tree/amazon-alexa

Build credit

Entrant: Jose Padilla AI engineering assistance: ChatGPT by OpenAI

Judge shortcuts

These files are included so a judge can reproduce the working path, inspect the product feedback and friction evidence, and verify the hackathon-window contribution without relying on claims in the project description alone.

Built With

  • agentic-ai
  • alexa+
  • codeql
  • github-actions
  • html
  • javascript
  • mcp-python-sdk
  • model-context-protocol
  • python
  • streamable-http
Share this project:

Updates