Forward-Deployed AI Engineer

1 нед. назад

125k–175k USD / yearUnited StatesCanadaWorldwideSeniorRemote

aimachine learningsoftware development

Engineer responsible for owning and developing internal AI production systems in a fintech company.

Обязанности

is a ~dozen-person YC fintech (S21) building banking and payments for construction. Real accounts, instant ledgering, money movement that has to be right.
We already run on AI in two places, and both need a full-time owner.
1. Internal AI systems that are already in production. We have started an internal AI harness. Our fraud and customer-support back office is built with AI: case work, review, the ops surface, not a chatbot bolted on the side. That system needs someone who maintains it, extends it, and treats it like a product, every day.
2. The engineering factory. The team already ships in Cursor / Claude Code. Setup is uneven. Venvs, VMs, skills, and repos drift. Skills rot. We do not have a daily picture of what shipped or how good the code is. You make every engineer dangerous in the same way.
You sit on the engineering team as a forward-deployed AI engineer . Embedded with the people doing the work. You own the harness and the back office as real software. You also own environments, team skills, repo contracts, quality scrape jobs, and process so the factory gets tighter every week.
If you only want to write product features, stop. If you only want to do "AI enablement" and will not own the back office, stop. If you want to build and run the AI layer the company already depends on, keep reading.
You have fewer than ~six years writing production software
Your AI story is ChatGPT in the browser and a LinkedIn course
You want to be a product IC who "also helps with AI," or you only want developer-productivity work and will not own the fraud/CS back office
You need a 40-person developer-productivity org, a manager ladder, and a Jira queue of enablement tickets
You will not show a harness or OSS artifact we can inspect
You need visa sponsorship we do not offer

Требования

6+ years as an engineer , shipping production software. Languages and stacks matter less than: you have been in the mess long enough to have taste.
Proof we can click: a harness, an eval framework, an agent/skill system, or a meaningful open-source contribution in this area. Link it at the top. "I use Cursor a lot" is not enough.
You have set up other people's environments, not only your own. Dotfiles-as-theater does not count. Repos that boot, tests that run, agents that have the tools they need.
You treat harnesses, skills, and the back office like production code: versioned, reviewed, deprecated when they rot. Comfortable in Python/TypeScript or similar: CI, scrapers, MCP/tool servers, sandboxes, ops UIs.
Small-team operator. You will not have a platform org. You are the function for internal AI and eng acceleration.
Must meet the location / work-auth rules in Sanity Check.

Будет плюсом

You have already run a shared Cursor / Claude Code / Codex (or similar) setup across a team
Published evals, SWE-bench-style loops, or sandbox/VM runners
Contributions to Hermes, Claude Code skills, Aider, Continue, OpenHands, or similar
You have shipped AI into an ops/fraud/support workflow, not only coding agents
You have opinions on skills vs memory vs evals, and scars to match

Условия

Canada remote: CAD 150,000 – 250,000. US remote: USD 125,000 – 175,000. Equity 0.25% – 0.50% . We can flex cash vs equity for the right person. The number is in the currency of the country you work from.
Remote in the US or Canada, as above
Real ownership of internal AI (harness + back office) and how the eng team uses it
Medical/dental/vision, 401(k), reasonable PTO, equipment
A small, sharp team. Zero bureaucracy. High leverage.

Как откликнуться

Send:
Resume and GitHub / LinkedIn
The proof first - harness, eval, skill pack, or OSS contribution. Link it.
A short note (half page) covering: six-plus years (what you actually shipped); one time you made other people faster (engineers or ops), not just yourself; what you would inspect first: our harness, the fraud/CS back office, or the Cursor/repo/skill setup
Optional: 3-8 minute Loom walking through the harness or OSS work (no secrets)
If the application reads like generic AI sludge, we will know. Name the repo. Name the failure mode you fixed.

Другое

Remote (US or Canada) · Full-time · Engineering · 6+ years Canada remote: CAD 150K – 250K · US remote: USD 125K – 175K · 0.25% – 0.50% equity WE ALSO OFFER IN PERSON IN AUSTIN TEXAS STARTING JAN 1 2026
Remote is this seat. US or Canada. TN is OK if you are Canadian. We are not sponsoring other visas. Pay is in the currency listed for that location: CAD if you are remote in Canada, USD if you are remote in the US.
You have been a working engineer for at least six years . This field is new. We do not expect six years of "AI engineer" on a resume. We do expect you have shipped real software for a long time, and that the AI/agent work is the reason you want this job.
Show the work. A harness, an eval loop, an agent skill pack, or a real open-source contribution we can click. Blog posts about "using ChatGPT" are not proof.
This is an engineering seat. You will own production internal AI systems (harness + fraud/CS back office) and accelerate the eng team. Not DevRel. Not enablement theater. Not a prompt intern.
Own the internal AI harness. It exists. You maintain it, harden it, and keep building it full time. Evals, tool wiring, sandboxes, whatever makes agents reliable here.
Own the AI back office. Fraud and customer support already run on an AI-built back office. You extend it, keep it correct, and sit with ops when a case flow is wrong. This is production, not a demo.
Make every engineer's environment actually work. Cursor, agent runtimes, virtualenvs, VMs, secrets, language toolchains. When someone clones a repo, the agent can boot, test, and ship. You notice when it cannot.
Own team skill sets. Shared skills for our stack and our way of working. You keep them current. Stale skills are a bug you fix, not a wiki someone might read.
Set the repos up for agents, not just humans. Devcontainers or equivalent, task runners, test entrypoints, lint, fixtures, docs the model can follow. The factory has a contract.
Build jobs that scrape the work. Daily signal on what shipped, review quality, test health, agent usage, where we are slow or sloppy. Improve the factory. Do not build a vanity dashboard.
Install process that sticks. How we open PRs, how we review with agents, how we promote a skill, how we kill a bad one. Boring, enforced, light.
Sit with the people who use it. Engineers and ops. You watch a session or a case fail and fix the harness, the skill, or the back office. You do not throw a Notion page over the wall.
15-minute founder screen
~1 hour working session on a harness, back office, or environment problem (we will look at your artifact)
Conversation with engineers you would sit beside
References
is a YC-backed (S21) fintech: banking and payments for construction. Companies deposit, collect, and pay out in one place. We are the bank contractors use, and a pay button inside construction SaaS.
We started with MazumaGo in Canada, learned our customers were construction, and built a US banking and payments engine for SMB trades. Soft-launched beta in 2023, fully launched in 2024, growing since.
Construction still runs on slow payments, paper checks, and fees that punish small businesses. We fix that with real accounts, instant ledgering, and money movement that has to be right. Engineering quality is not optional when the blast radius is money.
A note from the founder: I'm a technical founder. We already have an internal AI harness, and the fraud and customer-support back office is built with AI. The eng team already lives in Cursor. What I still need is someone who wakes up owning that layer full time: the harness, the back office, and the factory (environments, skills, repo contracts, quality loops) so ops and eng compound on the same system. If you have six-plus years of real engineering and you can show a harness or OSS work in this new field, let's talk.