You will architect and build the agent harness that powers Octavia's end-to-end critical minerals exploration platform. This is not a wrapper around someone else's framework — you will design the harness, the tool surface, the planner, the eval loop, and the production infrastructure underneath. Agentic systems are the entire product. You are the engineer who makes them real.
// You'll build
- The agent harness itself — the runtime that orchestrates LLM calls, tool use, multi-step planning, retries, parallelism, and long-running scientific workflows. This is the core product surface.
- The tool layer — typed, observable, replayable wrappers around our proprietary geophysics models, geospatial queries, seismic interpretation routines, drilling-program calculators, and external data sources.
- Memory and state — durable conversation, scratchpad, and project state so agents can run for hours or days across exploration workflows without losing context.
- The eval harness — replay-driven evals against ground-truth wells and historical discoveries; regression suites for every prompt and tool change; LLM-judge pipelines you actually trust.
- The platform underneath — backend services, APIs, frontend interfaces, cloud infra, data pipelines, model serving, and scalable compute. You own the whole stack until we hire behind you.
- The novel parts — when off-the-shelf agent patterns don't work for scientific data, you invent new ones, grounded in the physics and math of the problem.
- The engineering culture — best practices, code standards, on-call hygiene; the bar that future hires will be measured against.
// We're looking for
- You have built an agentic system end-to-end. Not a chatbot demo — a system that uses tools, plans, recovers from failure, and ships value to real users.
- Production AI/ML infrastructure experience. You have deployed real MLOps: data pipelines, model serving, eval suites, observability, version control for prompts and weights.
- Full-stack range. You are equally comfortable in backend architecture, frontend, and infrastructure. You ship across the entire stack without flinching.
- Strong systems language. Production experience in Go, Rust, or Java — plus the Python proficiency that agentic AI demands.
- Cloud-native. AWS (EC2, S3, Lambda), GCP (Vertex AI, BigQuery), or Azure. Docker, Kubernetes, CI/CD as second nature.
- Scientific or geospatial computing literacy. Data-intensive applications, irregular numerical data, real units, real noise.
- Builder's temperament. Resourceful, resilient, entrepreneurial. You offer solutions, not problem lists. You get dopamine from solving hard things. You are punctual, organized, and you ship.
- Founding-engineer instincts. You thrive in ambiguity and would rather invent the tool than wait for it. You can lead and mentor as the team grows.
// Requirements & bonus points
- Authorized to work in the U.S. (no visa sponsorship available)
- Master's degree
- 3+ years of software development experience
- In-person, Palo Alto, CA
- Bonus · Strong Python proficiency
- Bonus · Front-end framework experience (React, Vue, Angular)
- Bonus · Palantir Foundry / ADP experience
- Bonus · Prior early-stage startup experience
Ready to apply?
Tell us about an agentic system you've shipped. We read everything.