Agent Infrastructure (IC)

Welcome to Matter, where we are building the future of vision AI: pairing a world-first sensor that sees molecular chemistry, temperature, and 3D shape with a Large World Model that will be the most powerful intelligence engine for the physical world. This system doesn't just see what something looks like; it understands everything from a single pixel. We call this Superintelligent Vision.

Our team has delivered technologies to Mars for NASA/JPL, designed advanced sensors for U.S. Defense, and frontier artificial intelligence systems. We are now building the next generation of space- and airborne-based sensing systems.

Matter is hiring an Agent Infrastructure Engineer to build the production systems that turn physics-informed and general-purpose AI into secure, evidence-backed intelligence and action. Reporting to Ignacio Cases Martin, this hands-on individual contributor will work across agent runtimes, reasoning graphs, model routing, retrieval, memory, tools, durable execution, and human approval in partnership with research, platform, product, design, and evaluation teams.

• →

  Build and operate a production agent runtime that composes physics-informed models, LLMs, VLMs, world models, scientific code, data services, and deterministic business rules through stable interfaces.

• →

  Design durable execution with state, planning, tool selection, retries, timeouts, cancellation, checkpoints, provenance, human approval, and recovery.

• →

  Build retrieval, RAG, memory, and indexing patterns that respect tenancy, permissions, freshness, versioning, retention, and evidence lineage.

• →

  Create evidence-first outputs that distinguish source data, model inference, system-generated synthesis, user input, and unresolved uncertainty.

• →

  Integrate continuous evaluation across task success, groundedness, scientific validity, safety, latency, cost, and human intervention.

• →

  Partner with product, design, AI platform, Signal and Evaluation, and Telemetry teams to create reusable APIs, tools, and human-in-the-loop workflows without one-off orchestration.

To comply with U.S. export regulations, applicants must be one of the following:

• A U.S. citizen or national

• A lawful permanent resident (green card holder)

• Eligible to obtain required authorizations from the U.S. Department of State

• Experience building production agent infrastructure, applied-AI platforms, workflow engines, distributed systems, or complex AI products.

• Strong hands-on software engineering skills across architecture, APIs, stateful services, databases, asynchronous execution, testing, debugging, and production operations.

• Practical experience with LLMs, VLMs, multimodal systems, RAG, memory, tool use, model routing, structured generation, and common agent failure modes.

• Experience building systems with security, permissions, data isolation, auditability, and human approval requirements.

• Sound product judgment and the ability to explain uncertain system behavior precisely across technical and non-technical teams.

• Experience with scientific, geospatial, industrial, defense, autonomy, or other high-consequence customer-intelligence systems.

• Experience building multimodal applications that combine imagery, time series, maps, documents, structured data, or sensor streams.

• Experience with agent evaluation, reinforcement learning or world-model environments, model routing, constrained decoding, or durable workflow systems.

• Familiarity with edge or intermittently connected operation, local model execution, or mixed cloud and edge architectures.

• Production agents generate useful outputs with clear evidence, provenance, uncertainty, and human authority boundaries.

• Long-running workflows recover predictably from unavailable models, tools, data, or network connectivity.

• Reusable APIs, SDKs, tools, and graph components let teams create new agents without one-off orchestration.

This role is based in San Francisco, CA, and requires onsite work.