Senior Staff Engineer

Team: Engineering

Location: Remote (India), Bengaluru

Commitment: Remote

Workplace Type: remote

Where youll leave your mark?

Agents whose output nobody re-checks. Youll see how we prove an agent is trustworthy: evaluation harnesses that gate releases rather than run quarterly, golden sets built with real compliance experts, regression suites assembled from production traces, and drift detection that fires before a customer notices. Not a set of scripts — a platform every other team builds on. Youll know it works when a compliance lead stops reviewing a category of output, and nothing breaks.

A system that can prove what it said, and admit when it didnt know. Compliance is a claim about the past, which means state as of a date has to be held separately from when we learned it, and every artefact has to carry unbroken provenance from the source system to the audit package. The other half of the problem is honesty under failure. At several hundred connectors, partial failures are normal, and a connector returning nothing because Okta is down must never look like a control genuinely failing. Passing, failing and unknown are three different states, and that distinction has to survive all the way to the interface.

Beyond those two, there are surfaces youll work across without owning alone: tenant isolation and enterprise identity as we move upmarket, the inference economics that decide whether continuous monitoring is viable at scale, and how we model 200 frameworks so the next one is cheap. And since every engineer here will be building with agents within a year, well expect the practice to be better across the org because you joined.

The kind of builder were looking for -

Someone who has built something that changed the shape of the work. Not a prototype — a system where an LLM took autonomous, multi-step action against real data and real users, and displaced something people used to do by hand. Well want to hear what it replaced, what it cost, what its failure rate was, and how you measured it.

An engineer who treats the model as an untrusted subsystem. Constrained outputs, independent verification, thresholds that escalate rather than guess, a clear line between what the model decides and what it only proposes. Were more interested in how your system detected and contained a bad output than in how it tried to prevent one.

Depth in at least two of these, with room to pick up the rest here: production LLM and agent systems; temporal and provenance data modeling; high-fan-out integration reliability; multi-tenant and enterprise identity; inference cost engineering; domain modeling for messy regulatory rules.

10+ years building software, at least two of them at Staff scope — setting technical direction that engineers outside your reporting line followed.

Range. Some quarters this role is an architect setting direction, some quarters its the person taking on the problem nobody has cracked, some quarters its carrying a team through a hard delivery. The people who do well here read what the company needs and adapt.

Willingness to challenge how we build, and to hold the line when it counts. Were hiring partly for disagreement — on our architecture, our agent design, our standards. That includes telling go-to-market what we cant yet defend, and offering a path rather than a refusal.

Influence without authority. Nobody on this team reports to you. Your leverage is clarity, credibility and the quality of your reasoning across Engineering, Product, Security and CX — and since were remote-first, most of that happens in writing.