Work/Autonomous Agents

Closer: iMessage Negotiation Coach

iMessage coach that calls a used-car seller's bluffs with a Bayesian model

Typeexperiment
Statusprototype
Updated2026-07-25
SourcePublic

What it is

Hackathon project: a phone number you text a used-car listing. It researches the real value and tells you the exact number to counter with, using a Bayesian model of the seller's hidden price floor that updates each turn. There is no dashboard; every action is a message or a rendered card.

The problem

Private-party car buyers overpay because they cannot read seller bluffs or negotiate.

How it works

  1. FastAPI and NumPy backend running locally; Linq delivers iMessage webhooks to localhost over an outbound connection, so no public URL is needed
  2. A belief curve over the seller's floor moves when the engine detects a bluff_claim signal such as 'three other buyers are coming Saturday'
  3. Runware handles all AI inference; the user is identified by the phone number on the inbound webhook, with no signup or login
  4. Multi-deal switching, lifetime-savings tracking, and a Makefile preflight and runbook so the number never silently goes quiet during a demo
  5. Demoed against the Marcus seller agent as a live counterparty

Stack

PythonFastAPINumPyLinqRunwareiMessage

Disciplines

For your business

Conversational agents that negotiate or advise over SMS and iMessage with explicit, inspectable decision logic.