Open role

Member of Technical Staff, Evals

Remote · Full-time

Build the deterministic runtime and evaluation architecture behind Strust's behavioral feedback loop.

About Strust

Strust is building the behavioral eval layer for software changes: deterministic feedback for coding agents and clear behavioral diffs for human reviewers.

We started with COBOL-to-Java migration because a running legacy system is an unusually strong reference. The larger mission is to help every team understand what agent-written software actually did before they trust the change.

The role

The harness runs software under controlled inputs, captures the behavior that matters, and exposes the first divergence in a form an agent can act on and a human can review. Today it goes deepest on paired COBOL-to-Java execution.

This is infrastructure work with a research edge: the interesting problems are what to measure, how to replay it deterministically, and how to preserve human understanding as agents write more of the code.

What you'll do

Build the deterministic runtime that controls inputs, time, and side effects for repeatable evals.

Own the loop from trace capture and differential execution through agent feedback and human review.

Return the smallest useful failure and first divergence to coding agents.

Make behavior changes legible to reviewers without reducing them to a single score.

Turn confirmed failures into evals that run on every later change.

Take the harness beyond COBOL-to-Java while keeping migration as the first deep lane.

Work directly with the founder on the product and its technical direction.

You might be a fit if

You write strong Python (JVM familiarity helps) and have built production or research infrastructure that other people depended on.

You have operated distributed systems, data pipelines, or test infrastructure that had to be reliable at scale.

You care about behavior at the level of rounding modes and branch semantics, not just green CI.

You find differential testing, metamorphic testing, program analysis, or eval harnesses genuinely interesting.

You communicate clearly in writing and optimize technical systems for their readers.

Nice to have

COBOL, mainframe, or legacy-modernization exposure.

Compilers, static analysis, or symbolic execution background.

Experience building LLM eval harnesses or LLM-assisted code translation tooling.

What you get

Founding-engineer equity and a say in everything.

One concrete mission: make software behavior easier to test and understand.

Direct customer exposure while the product is still taking shape.

Apply

Apply for Member of Technical Staff, Evals

Tell us what you have built and what made the work matter. We read every application.