Software Engineer (Remote)

Who we are

Keye is building the diligence layer private equity has always needed and never had. Today it does the work of an analyst, at the bar of an investment committee, on the clock of a two-week deal — 80% of the work in under five minutes, against a process that otherwise takes weeks. Firms use it to kill bad deals earlier and save millions in third-party diligence costs. Tomorrow it watches the deal after you sign. Benchmarks it against every comp in the market. Alerts you before the thesis breaks.

Not a faster analyst. A second brain with a hundred years of scars and a live pulse on every industry it touches — built by a team that's been on the other side of the table. Four acquisitions as early employees. Twenty-plus years in M&A investing. Engineers out of Tesla, Goldman, Vista, and McKinsey, with technical DNA from AI robotics, drug discovery, and ecommerce.

Top-10 PE funds already run diligence on Keye. $1.4T+ in AUM on the platform. 600% growth in six months. Backed by Sorenson Capital, General Catalyst, and Y Combinator.

What you’ll do:

You'll build the AI systems top PE funds run live deals on. Ingestion that structures chaotic data rooms. Agents that run deterministic analysis on top. Review surfaces that make model output trustworthy enough for an investment committee. Evals, retrieval, tool use, orchestration — held to the bar of a real deal clock.

You'd be joining a tight engineering team where everyone is load-bearing. One week you'd be heads-down on architecture — the systems that turn raw deal files into investor-ready outputs at 100% accuracy. Another week you'd be prototyping an agent harness nobody's built before. Another week you'd be cleaning up something fragile because a top-10 PE fund is running a deal on it tomorrow morning.

Ownership here is rotating, not absent. At any given time you'll be the person accountable for some load-bearing piece — a system, a migration, a fix — and you'll hand it off or expand it as priorities shift. You'll sit in on planning, align on sequencing, and have weight in what the next two weeks look like. Every shortcut compounds. Every good abstraction compounds harder. Customer obsession matters — their timelines, their standards, their refusal to accept 99%. The work is downstream of that pressure. You'll build things that have to meet it.

Example duties

Deepen the agent layer. We view the agent as a first class citizen. Not a demo. This means harnesses, planning, subagents, compaction, tool-use design — the primitives that decide whether the system scales with complexity or collapses under it. This will require iteration, decisiveness, and staying abreast on frontier updates.The faster this layer moves, the faster the product gets sharper.

Own the deterministic core. Our computation and fact gathering allows any consumer to pull any intermediate artifact from a deal, compose real analytical logic on top, and return a number an IC would defend.

Build evals that actually catch regressions before customers do. Not vibes. Real test harnesses for agent output, extraction accuracy, and analysis correctness.

Harden the ingestion layer so it handles data rooms with unknown schemas, malformed files, and edge cases that only show up in real deals.

Own the infrastructure that keeps the platform fast under real load. Concurrency, queueing, caching, cost — the systems work that decides whether we can run ten deals in parallel or ten thousand.

Location

Hybrid or Remote (North America).

What we're looking for

• You think in systems and ship in days. You know when to build the abstraction and when to skip it.
• You've succeeded at something hard and can point to it. A company you started. A team you led. A product people used. A bar you cleared that most didn't.
• You've built and operated production systems that broke, and you fixed them. Not prototypes. Not side projects. Real consequences.
• Broad across the stack. No layer is someone else's problem.
• Languages/tools: Python/React sho