Embed with general contractors and specialty trades. Model the job from bid to closeout and deploy the agents that take the paperwork off the field.
Full‑time Remote (US) Regular travel to client sites Reports to the founders
Construction is one of the largest industries in the country and one of the least digitized. Estimates live in spreadsheets, RFIs and submittals in email, job cost in an ERP nobody trusts, and the real schedule in a superintendent's head. A forward‑deployed engineer on this track embeds with general contractors and specialty trades, models how a job actually runs from preconstruction to closeout, and deploys the agents that take the paperwork off the field.
This is a new vertical for Topsail with the first engagement scoped from bid to closeout. You would be building the playbook, not following one.
What you would build
- Estimating and bid agents. Read plans, specs, and historical job cost to produce a first‑pass estimate and a compliant bid package, with the estimator reviewing instead of retyping.
- RFI, submittal, and change order triage. Agents that read the incoming email and documents, route them, draft the response, and keep the log current without a project engineer doing it by hand.
- A job cost ontology. One model of jobs, phases, cost codes, vendors, and commitments across the ERP, the accounting system, and the field, so the PM and the CFO are looking at the same number.
- Closeout automation. Punch lists, O&M manuals, warranties, and lien waivers assembled from what the job already produced.
What you will do
- Run discovery with executives and frontline staff, and turn a vague mandate into a ranked roadmap with baseline metrics the sponsor signs off on.
- Model the business and install the ontology: the entities, relationships, and rules the company runs on, resolved from spreadsheets, email, ERPs, CRMs, and the things that only live in people's heads.
- Design and ship the agents, retrieval systems, and automations that run on that foundation, model‑agnostic by default so the client can swap models without rebuilding.
- Own evals and observability for what you ship. Nothing counts until it is measured against the baseline and people are using it.
- Train client teams to use what you built, and to keep building without you.
What we look for
- You have shipped a production system that real users depend on. Not a demo, not a notebook. We will ask you to walk us through it in detail, including the parts you would do differently.
- You think in data models. You have designed schemas for a real business domain, done entity resolution on ugly data, and know where an ontology earns its keep and where it is overkill.
- You build with agents for a living. Agent SDKs, retrieval and chunking strategy, hybrid search, evals, and the tradeoffs of a deployment stack are things you have opinions about because you have been burned by them.
- You live in Claude Code, Codex, Cursor, or similar every day and can show us how.
- You know construction, or you have built for an industry like it. Field operations, job costing, subcontractors, and the gap between the office and the site. If you have swung a hammer or run a job, say so.
- You are comfortable in a trailer and a boardroom in the same day, and you write clearly enough that a superintendent and a CFO can both follow the plan.
Nice to have
- Time at a general contractor, a specialty trade, or a construction software company.
- Familiarity with Procore, Sage, Viewpoint, Bluebeam, or the estimating tools around them.
What you get
- A client relationship and the systems behind it, owned by you from week one.
- A new operating problem every few months, next to the people who built the company.
- Remote‑first, with time in New York, San Francisco, Boston, and Lehi, Utah when the client work calls for it.
- A serious AI budget: frontier model access, the tools you want, and a lot of tokens.
Compensation: Above‑market salary and bonus, plus meaningful equity.