5+ years of experience in building and operating distributed systems critical for businesses.
Proficient in at least one programming language: Python, Java, Go, C++, Rust, or TypeScript.
Experience shipping LLM or agentic systems that are actively used in production.
Strong background in backend services, APIs, and client architecture for product development.
Comfortable navigating ambiguity and shaping platform decisions early in the process.
Responsibilities
Own the design and implementation of agentic products or AI infrastructure end-to-end.
Build AI-native tools that operators use daily, focusing on full-stack development.
Serve and operate models in various environments, ensuring fine-tuning and data access.
Develop evaluation and observability systems to monitor performance and quality.
Rapidly transition user problems into production-ready systems.
Benefits
Opportunity to work on impactful projects that ship to real businesses quickly.
Flat organizational structure promoting direct collaboration and influence.
Engagement with cutting-edge AI technologies and infrastructure.
Chance to shape the future of the platform and its capabilities.
Full Job Description
About the Role
Members of Technical Staff (MTS) build the platform that everything else at Beacon runs on. You will own a piece of it end-to-end: design, implementation, operations, and long-term technical direction. This is a Staff Engineer role in everything but name. We run flat.
We are hiring for two roles. The bar is the same; the specialty differs.
Product and platform engineering. A senior engineer who builds agentic products and the platform underneath them: services, APIs, data access, and deploy infrastructure. Works directly with the people who use them.
AI infrastructure. An engineer who builds the systems around models: serving, retrieval, evals, observability, fine-tuning, and deployment into secure and regulated environments.
Both seats ship to real businesses, fast. You take a problem from the first conversation with a user to a system in production, and measure success by whether people use it.
What You'll Do
You will own one or both of these areas end-to-end:
Agentic products. Build the AI-native tools that operators across our portfolio use every day. You own the full stack: agent loops, UI, internal APIs, and the deploy infrastructure other engineers build on. The bar is not demos. It is tools people choose to open on Monday morning.
AI infrastructure. Serve and operate models, including open-weight ones, across cloud, VPC, and isolated environments. Fine-tune open-weight models on proprietary data where a general model falls short. Build the retrieval and governed data access that ground agents in real business data. Choose which model handles which job, with cost, latency, and quality as explicit trade-offs.
Evals and observability. Build the harness that tells us whether the system is working: task-level evals from real workflows, traces and telemetry, replay, and regression suites that catch quality drops before users do. Use the same signals to speed up root-cause analysis when things break.
How you will work:
Ship to real businesses, fast. Take a problem from the first user conversation to a system in production.
Prove quality with data. No model or prompt change ships on vibes.
Work agent-first. Agents are part of how you build, not just what you build.
Who You Are
Deep systems engineering. 5+ years building and operating distributed systems that real businesses depend on: databases, data platforms, query engines, control planes, or large-scale services. You have profiled your way to a 10x win, and you know where things break: partial failure, capacity limits, multi-tenant isolation, schema and API evolution. You write clean code in at least one of Python, Java, Go, C++, Rust, or TypeScript.
LLMs in production, not in slides. You have shipped LLM or agentic systems that people rely on: RAG over real enterprise data, agentic workflows, LLM-assisted diagnosis or operations. Coding agents are part of your daily loop. You know how these systems fail: hallucinated tool calls, silent quality drift, cost blowups, evals that measure the wrong thing.
Close to the customer. You have been on customer calls, run escalations, and turned what you heard into a product. You measure success by adoption, not launch.
Built from zero, run in production. You have been a founding engineer or taken a system from 0-to-1 through GA. You have also carried the pager: incident command, postmortems, and the fixes that keep the same outage from happening twice.
Comfortable with ambiguity. The scope of the platform is still being defined. Decisions you make in week one will shape what is possible in year three. That should energize you.
For the product and platform seat: strong on backend services, APIs and client architecture, internal tools, and deploy infrastructure, with enough front-end range to ship a full product. You have built products where the agent is the core of the experience, not a chat box bolted on.
For the AI infrastructure seat: hands-on with model serving (vLLM, llama.cpp, or similar), retrieval and vector search, evals, fine-tuning, and model routing. You have deployed into environments you do not fully control, such as customer VPCs or regulated and isolated regions, and you reason about GPU memory, batching, and quantization as fluently as about API design.
Preferred Skills
Built on or inside modern data platforms: Iceberg, lakehouse architectures, Snowflake, Databricks, or data catalogs and governance.
Performance engineering at scale: profiling, capacity planning, and measurable cost reductions.
Shipped into regulated environments: FedRAMP, government or sovereign regions, or air-gapped deployments.
Built LLM-assisted observability or root-cause analysis over real production telemetry.