About this role
About the position
This role lives where research meets infrastructure: you'll form hypotheses, test them fast with hands-on analysis, and then build the production systems that put the winning ones to work. LLMs are a core part of how we build here: a first-class tool, not a side experiment. The engineers who thrive are the ones who can drive an agentic harness as naturally as they read a profiler, and who know precisely where each of those tools stops being trustworthy.
Responsibilities
• Grow our event-driven approach across exchange feeds, market news, social platforms, and infrastructure telemetry.
• Stand up quick analyses in Python/Jupyter to gauge signal quality, then convert what you learn into concrete system improvements.
• Design and operate agentic harnesses: tooling, context management, evals, guardrails (that do meaningful work against our data and infrastructure, and own their quality once they're live).
• Deploy LLMs where they truly earn their place (extraction, classification, triage, faster research) and knowingly skip them where they don't.
• Own performance end to end, from the network edge through in-memory stores — instrumenting, monitoring, and debugging live systems shoulder to shoulder with operations while keeping SLOs tight.
• Drive green-field builds, design reviews, and post-mortems.
Requirements
• Three to five years building real-time or data-intensive systems (we care more about depth and trajectory than the precise year count).
• Real depth with LLMs rather than surface familiarity; how these models actually behave (context windows and their failure modes, tool use, structured output, cost and latency trade-offs, keeping hallucination in check) and how to construct the scaffolding around one: tools, memory, retries, evals, and sensible human-in-the-loop limits. Come ready to walk us through something real you shipped, end to end, including what went wrong.
• Solid engineering fundamentals in Python, Go, or Rust. That's a preference, not a gate: deep systems experience in another serious language carries over.
• Strong grasp of network programming and protocols: TCP/UDP/IP, DNS, BGP, HTTP(S), WebSocket, QUIC.
• Taking rapid POCs to production while confirming statistical significance, iterating quickly, shipping, and explaining the outcome clearly to technical and non-technical audiences alike.
• A Bachelor's or Master's in Computer Science, Data Science, Mathematics, or a comparable field.
Nice-to-haves
• You want problems that are truly hard and measurable, where the scoreboard is real and visible inside the team.
• You move quickly without cutting corners, and you'd rather ship, measure, and adjust than deliberate indefinitely.