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.
Qualification
Practical experience with LLMsExperience with scientificExperience with agent evaluationWhat Success Looks LikeReusable APIs
Required
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.
Preferred
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.
What Success Looks Like
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.
