Senior Product Manager, RL Environments — Handshake AI

Handshake•San Francisco, CA

8d•$180,000 - $220,000•Hybrid

About The Position

Handshake AI builds the training data that frontier labs use to push their models forward. A growing share of that work is in reinforcement learning: realistic, end-to-end environments where models can be evaluated and trained against real-world workflows. We’re hiring a Senior Product Manager to own the product surface that turns environment creation from a bespoke, weeks-long lift into a repeatable factory. Today, building a single RL environment is a substantial cross-team effort involving dozens of manual steps across operators and engineers, and depends on tribal knowledge across data sourcing, de-identification, synthetic data generation, tool building, packaging, and quality assurance. Frontier labs are asking for environments across many verticals at once, and the manual model doesn’t scale. Your job is to make it scale: design and ship the platform that compresses lead time, replaces hand-built workflows with self-serve tooling, and lets a small team of operators turn out high-quality environments for any vertical our customers prioritize. You’ll sit at the bridge between Operations and Engineering. Operators are running data pipelines locally, manually de-identifying datasets, manually QA’ing tools, and chasing customer deliveries. You’ll translate that work into a product roadmap, partner with our engineering leads on architecture and execution, and keep our research, GTM, and customer-facing teams aligned on what good looks like. This is a high-leverage, 0→1 role inside a fast-moving research-adjacent product space. You’ll work cross-functionally with Forward Deployed Engineering, Operations, Research, Design, and GTM, and your work will directly determine how many environments Handshake can ship, how fast, and at what quality bar.

Requirements

5+ years as a product manager shipping production code with engineering teams, ideally on platform, infrastructure, or developer-tools products.
A track record building tools for internal users (operations, forward deployed engineers, data teams, or similar) where reducing manual work and supporting power users is the core of the job.
Comfortable owning a product surface with many moving parts and dependencies (data pipelines, environment runtimes, tooling, QA, packaging) and sequencing roadmap work to unblock the biggest bottleneck next.
Strong product instincts in ambiguous, fast-iterating spaces. You can scope a problem from a Slack thread to a PRD to a shipped feature in hours, not days, and you don’t wait for a rigid spec before moving.
Data-informed and action-oriented. You use lead time, throughput, quality scores, and operator pain to prioritize, and you move quickly once the signal is clear.
Comfortable acting as the connective tissue between Operations and Engineering, translating ops pain into engineering work, and engineering tradeoffs back into ops-readable plans.

Nice To Haves

Background in reinforcement learning, frontier model training data, evaluations, or model post-training workflows.
Experience as a PM on developer tools or developer platforms, where the bar for power-user UX is high.
Familiarity with data pipelines, de-identification, synthetic data generation, or large-scale data QA.
0→1 experience standing up new product surfaces inside a fast-moving research-adjacent org without a fully formed playbook.
Experience at a marketplace, gig platform, or human-data company; or exposure to AI/ML data pipelines or annotation workflows.

Responsibilities

The Environment Factory. The end-to-end product experience for building and shipping an RL environment. Today this is a manual playbook; you’ll define and ship the platform that lets operators run many environments in parallel, with most steps in-product rather than off-platform.
Tooling, packaging, and delivery. Drive the roadmap for the tool registry, environment packaging, and customer delivery so labs receive a portable, deployable environment that runs reliably in their own infrastructure. Reduce time-to-deliver and the rate of last-minute rework on the day of delivery.
Quality at the frontier-lab bar. Own the leveling framework for environment quality (currently L1–L5 by vertical and persona) and the roadmap that gets priority verticals from L1 to L4+. Define and ship the QA tooling that turns environment, task, and rollout QA from a manual review into a productized check.
Operator tooling. Operators are your primary users. Build the dashboards, in-product workflows, and self-serve flows that replace the manual work they do today from data transformation to environment QA to delivery cutoffs.
Goals and metrics. Define and track targets including: environment lead time, environments delivered per quarter, % of in-platform vs. off-platform steps in environment creation, environment quality level by vertical, QA pass rate on environments/tasks/rollouts, tool registry coverage, and operator time per environment.
Cross-functional partnership. Work with Engineering on architecture and execution, with Operations on workflow and pain points, with Research on what environments need to support post-training and verifier work, with Design on operator UX, and with GTM and customer teams on what verticals to prioritize and how to package what we build.