Founding AI Engineer

Location: Remote (US) · Type: Full-time · Reports to: CPIO

About ContractorHUB

ContractorHUB is the operating system for home-services contractors — roofing, siding, gutters, windows, HVAC, solar. We aggregate the half-dozen disconnected tools a typical contractor runs (CRM, job management, accounting, call tracking, ad platforms, satellite measurement) into a single multi-tenant SaaS, then layer agentic AI on top so an owner, salesperson, or office staffer can run their day in plain language instead of clicking through eight tabs.

We are early. The platform is live, real contractors pay for it, and we have a deep bench of domain context — but the AI layer is still being built. The person we hire here will own that layer end to end.

The role

You'll be the founding AI engineer. That means you own the agentic platform, the ML pipelines, and every model decision we make for the foreseeable future. You'll work directly with the founder, ship to production weekly, and have an unusually direct line from "I noticed this in a customer call" to "I shipped a fix today."

Concretely, you will own:

1. The agentic platform

We already have a chat system with semantic tool retrieval (pgvector + OpenAI embeddings), a registry of tools spread across the app, and ~12 contextual "FAB" chat surfaces that adapt to the page the user is on. You will:

• Extend the tool registry as we open up new domains (inventory, EOS/L10 meetings, financials, marketing attribution).

• Improve tool-selection quality — we currently use cosine similarity over text-embedding-3-small; you'll evolve this toward better retrieval, reranking, and multi-step planning.

• Build the eval harness. Right now we have no systematic regression coverage for prompts or tool choice, and that is the single biggest risk to quality as we scale.

• Drive cost down. LLM spend is a meaningful line item; model selection, caching, and prompt design matter.

2. Computer vision for roofing

We're building a model that estimates roof measurements from satellite imagery. Phase 1 (data collection) is shipped — every satellite-aligned project automatically captures a labeled training sample. Phase 2 (segmentation) and Phase 3 (measurement estimation) are open. This is a real ML problem, not a wrapper around a foundation model — you'll choose the architecture, run training, and ship inference into the product. Roof measurement is currently a ~$50/report SaaS market that we'd like to disrupt internally.

3. Realtime voice

We have a plan to add OpenAI Realtime voice to every chat surface in the app — contractors are often on a roof or in a truck and typing isn't an option. The plan calls for a shared session service, a WebRTC client, and per-context tool exposure. You'll lead the build.

4. Conversational BI

The Insights page is the "ask your business a question" surface — chat-to-chart, chat-to-map, click-to-drill, save & pin. It already ships ECharts + deck.gl renderings driven by tool calls. We want to extend it toward predictive analytics, morning briefings, and insight-to-action handoffs (the chart suggests an action; one click executes it via an agent).

5. The boring, important stuff

• Evaluation, observability, and cost dashboards for every AI surface.

• Multi-tenant safety: tools must respect tenant isolation, role permissions, and PII boundaries.

• Documentation patterns so the rest of the team can add tools without breaking things.

What we're looking for

You probably have:

• 3+ years shipping production ML or AI systems (not just prototypes; not just notebooks).

• Deep, current familiarity with LLM patterns: tool calling, function schemas, RAG, retrieval evaluation, agent loops, structured output.

• Hands-on experience with at least one of: computer vision (segmentation/detection), realtime/streaming systems, or large-scale evaluation infrastructure.

• Comfort owning a system end to end — model choice, training, deployment, monitoring, cost.

• A bias toward shipping. We move weekly, not quarterly.
  

You will need to be willing to:

• Learn the home-services domain. Sit on customer calls, ride along on a roof inspection, read a JobNimbus export. The best AI features come from people who actually understand what a contractor's Tuesday looks like.

• Cross language boundaries. The application is Laravel/PHP; ML training can be Python; chat tools live in PHP. You don't need to be a Laravel expert on day one, but you'll be reading and writing PHP regularly.

• Make calls without a committee. There is no AI team to consult — you are the AI team. We'll back your decisions; you'll own the consequences.
  

Bonus:

• Background in geospatial / satellite imagery / photogrammetry.

• Experience with WebRTC, streaming audio, or realtime voice agents.

• You've built an eval harness for an LLM product before and have opinions about it.

• You've worked in a vertical SaaS company before and understand why "horizontal AI" doesn't always win.

Why this role is unusual

• Founding equity. This is a meaningful equity grant in a real, revenue-generating company — not a token sliver.

• Direct ownership. No layers between you and the product decisions. No AI roadmap dictated by a PM you've never met.

• Real data, real impact. Every model you ship affects a real contractor's revenue the same week. The feedback loop is days, not quarters.

• Greenfield. Computer vision for roofs, realtime voice for field teams, agentic CRM workflows — none of these have a clear winner in the contracting space yet. We have a head start and the data to keep it.

• Sane scope. This isn't a research role and we won't ask you to invent a new architecture. It's an applied AI role with hard, valuable problems.

How to apply

Send a note to dev@contractorhub.app with:

1. A short description of an AI/ML system you've shipped to production, and what the hardest decision was.

2. A link to anything we can read — a paper, a repo, a blog post, a project writeup. Quality over quantity.

3. Any thoughts on what you'd build first if you joined.
   

We respond to every application.