Agentic Build Engineer

Contract · 6 months · Full-time · Remote (work anywhere in the world, on US Eastern hours) · Paid in USD

Two seats. One company building software that has to be right.

What we're building

NexTrial.ai builds the intelligence layer that makes clinical trial startup predictable. We call the category Trial Activation Intelligence. In plain terms: we turn clinical protocols and regulatory requirements into verified, audit-ready outputs, so trials can start in weeks instead of months.

This is infrastructure, in a regulated domain. Two things follow from that. Being right matters more than being fast. And the system has to prove its own work, with every output traceable back to its input. If you have ever wanted to build software where correctness is the product and not an afterthought, this is that.

The two seats

Same core profile, different center of gravity. Tell us which one fits you, or that you can cover both.

Seat 1, Application and Orchestration. You build the agent chain and the data flow at the heart of the platform. Python, async, agent frameworks (LangGraph or similar), Postgres and pgvector.

Seat 2, Platform and Verification. You own the cloud environment and the test harness that proves the system works. Google Cloud (Cloud Run, Cloud SQL, Secret Manager, Artifact Registry), Terraform, Docker, CI/CD.

How we work (read this part twice)

This is the part that makes or breaks the fit. We do not build the way most teams build.

Spec-driven development (SDD). We write the specification first. The spec is the durable asset and the source of truth; the code is downstream of it and, honestly, disposable. You read the spec before you write anything, and when the spec is right, the implementation follows. If "just start coding" is your reflex, we are not a match.

AI-native, with you in command. We build with modern AI coding agents as real build capacity (think tools in the Claude Code family). You are not replaced by them, you direct them. You write the intent, review what comes back with a critical eye, and catch the mistakes they make. The skill we value most is not how much code you write by hand. It is how well you drive the tooling and verify the result.

Discovery before build. You read the codebase as ground truth before you scope or estimate. Our standing assumption is that more is already shipped than you think. Confirm reality first, then build.

Honest engineering, no exceptions. When something is genuinely broken, you diagnose it to the root and you fix it or flag it. You do not fake a passing test, hide a shortcut, or weaken a check to keep moving. In a domain where the system has to prove its own work, that instinct is the entire job. This is the one thing we do not compromise on.

Focused and async. Clear specs, clear stopping points, one thing at a time. We move fast by being deliberate, not by being chaotic.

Who you are

• Strong in Python. You write clean code and you read other people's code comfortably.

• You have code we can look at. A public GitHub, open-source contributions, or a portfolio of real projects. We want to see how you build, not just read about it.

• Genuinely fluent with AI development tools. Not "I tried it once," but you actually build this way and have opinions about it.

• You read before you change. Curiosity about the existing system comes before the urge to add to it.

• Comfortable working from a written spec with defined checkpoints, independently, across a time zone.

• English-speaking, able to collaborate clearly in English in writing and on calls. This is non-negotiable. The whole team and all our specs are in English.

• Startup mindset. You own outcomes, you are comfortable with ambiguity, and you would rather ship something real and verified than wait for perfect conditions.

• For Seat 2: solid Google Cloud and infrastructure-as-code experience.

Nice to have: document parsing, model serving or LoRA work, or any background in a regulated field (health, finance, anything with audit trails).

Our people and culture

We are small, the bar is high, and the work is real. We have traction, a clear category, and a way of building we are proud of. Four things we actually value, not poster words:

• Clear thinking. Say what you mean. Reason in the open.

• Honest systems. The system tells the truth about what it did. So do we.

• Execution under constraint. Limited time, limited budget, real output. That is the game, and it is a good one.

• Collaboration over ego. The best idea wins. Nobody defends a weak one to save face.

If that is already how you work, you will feel at home here fast.

Not a fit if

• You let AI write code you do not read or understand.

• You start typing before you understand the system.

• You treat correctness, tests, and audit trails as bureaucracy to route around.

• You need a big team, a manager checking in hourly, or fully specified tickets with zero ambiguity.

Logistics

• Contract, 6 months, full-time, remote. Work from anywhere in the world.

• Hours are the real requirement. You need substantial daily overlap with US Eastern Time (ET), at least 4 to 5 hours of live overlap. We work async, but this is high-touch, so we need real shared time, not just handoffs.

• If you are outside the Americas (for example India or Pakistan), that overlap means working an evening schedule on your side. Many of our best candidates do exactly this. We just want you to know the hours before you apply.

• Paid in USD.

• Start as soon as we find the right two.

• The first week is a paid trial on real work. It is how we both find out if this is a fit, with no guessing.

How to apply

Show us your GitHub. That is what we most want to see, more than a CV. Point us to a repository, a project, or a commit history you are proud of. A CV is welcome, but your code tells us more.

Tell us which seat fits. And in a line or two, tell us the most interesting thing you have built with AI tooling, and what you did versus what the tool did.

If you know someone anywhere who builds like this and can work Eastern hours, send it their way. The right two are out there.