Executive Summary

Executive Summary

The HVACOps AI copilot is a job-scoped assistant designed for technicians working under real-world constraints: incomplete data, safety-critical workflows, and strict time pressure. The product goal is not to answer questions, but to support a decision in the field with evidence-based responses and a clear audit trail.

This dossier outlines how we move from discovery to a pilot-ready system in four weeks, while keeping trust, security, and reliability as first-class requirements. The system design is Cloudflare-first: Workers orchestrate requests, D1 stores structured job context, Vectorize and R2 handle retrieval and artifacts, and AI Gateway mediates model calls with logging and caching.

Define the job: decision support inside a job-scoped workflow.
Trust posture: evidence-based answers or abstain.
Cloudflare-first backend for low latency and predictable cost.
Auth and tenancy boundaries enforced via dedicated Auth Worker.
Evaluation built into the delivery plan from week two.

Who this is for

  • HVAC owners and dispatch: fewer callbacks, better documentation, faster resolution.
  • Technicians: fast context assembly, safety-aware reminders, and evidence-backed summaries.
  • Security reviewers: explicit tenant isolation, least-privilege services, and auditability.
  • Engineering leadership: a clear scope, risk register, and delivery cadence.
Decision
We will ship a job-scoped copilot first, not a general HVAC chatbot, to keep context, safety, and ownership explicit.

What success looks like

  • Accuracy: 95% of answers grounded in retrieved evidence.
  • Latency: P95 response under 2.5 seconds for a typical query.
  • Adoption: 50% of technicians use AI copilot weekly.
  • Safety: zero unsafe suggestions in the high-risk eval set.
Tradeoff
We prefer determinism over creativity: temperature stays low, and structured output is enforced.

Non-goals

  • No autonomous troubleshooting without evidence.
  • No wiring diagram interpretation without a manual reference.
  • No cross-tenant insights or aggregate analytics in the copilot response.
Open questions / next steps: confirm the initial success metrics with stakeholders, prioritize risk categories (electrical, gas, refrigerant), and validate the minimum viable eval set.