platform engineer for internal AI systems

Supabase provides a Postgres development platform with an integrated backend solution that includes Database, Auth, Storage, Edge Functions, Realtime, and Vector Search. Its globally distributed, open-source-first team builds tools for developers and supports the open-source ecosystem.

• Build and ship the execution platform for internal AI agents, including event-triggered queues, a model-agnostic runtime, durable state, human review gates, atomic rollback, and complete run logging
• Build the evaluation layer with golden suites, behavioral assertions, written judge rubrics, safety cases, and CI gates that block regressions
• Build and register agents for reporting, drafting, linting, triage, and question answering across executive, team-lead, and individual-contributor levels, as well as agents that monitor and improve the platform
• Enforce governance through risk tiers, least-privilege credentials, tool-permission gates, decision audit logs, and autonomy classes that make forbidden actions impossible
• Design systems for contacting people with per-person interruption budgets, bundled messages, and useful information provided before requests
• Own platform tooling, including the compiler and validator, inventory integrity, and distribution of context and capabilities to repositories and chat surfaces
• Compute team maturity grades from production data through a pipeline that supports query-based review of disputed results
• Instrument the platform’s return with a ledger that tracks work absorbed by agents and calculates monthly value

• Have shipped production LLM agent systems used by others, with operational history, real users, and at least one incident to discuss
• Design evaluations using golden sets, behavioral assertions, judge rubrics, pass thresholds, and CI gates
• Have deep API experience with systems where work happens and have authored MCP servers
• Own infrastructure end to end in Python on GCP, including a cloud warehouse and infrastructure as code; provision, deploy, monitor, and roll back systems, and make architecture decisions
• Use agentic tools on real work in files and repositories, and be able to describe triggers, permissions, human approvals, logging, and a failure
• Be able to design software notifications with consideration for limited user trust
• Будет плюсом: public work in the field, such as an open-source agent framework, an MCP server, an evaluation harness, or cited writing on agent reliability; LLM observability and cost instrumentation; experience building an internal platform voluntarily adopted by non-engineers

• Fully remote; the company hires globally
• WeWork membership or co-working allowance usable worldwide
• ESOP equity ownership
• Tech allowance for work equipment
• Health insurance covered at 100% for employees and 80% for dependents, wherever they are
• Annual company off-sites lasting one week
• Flexible schedule in an asynchronous work environment
• Annual education allowance for courses, books, conferences, or other learning