Member of Technical Staff - Code Generation

embedding-vc; San Mateo (CA); On-site; USD 120,000 - 160,000; Full time

Job summary

A technology firm specializing in AI is looking for an Applied AI Research Engineer to design agentic systems and build evaluation frameworks for multi-step coding tasks. This role requires expertise in supervised fine-tuning and ensuring security in development environments. The company is committed to maintaining an on-site team in San Mateo, California, offering a collaborative work environment.

Qualifications

• Experience with multi-step coding and full-stack LLM engineering.
• Ability to design agentic systems and evaluate task suites.
• Proficient in supervised fine-tuning and safety filter implementations.

Responsibilities

• Design agentic systems and tool catalogs.
• Build evaluation frameworks for multi-step coding.
• Ensure security and isolation within development environments.

Skills

Agentic systems design

Evaluation mindset

Security & isolation

Supervised fine-tuning

Dataset curation

Job description

Introducing Moonlake, AI for creating real-time interactive content

Mission: As an applied AI Research Engineer: Code agents (post training + systems)

Scope of Work

• Agentic systems design: Tool catalogs, function calling, program synthesis/repair loops, ReAct/Reflexion/ToT/LangGraph-style control, self-verification, sandboxed execution.
• Evaluation mindset: Build task suites for multi-step coding Full-stack LLM engineer: Prompt libraries, routing, retrieval, KV-cache management, streaming, telemetry.
• Security & isolation: Docker/firejail, network egress controls, secrets hygiene, dependency pinning, supply‑chain sanity.
• Strong post‑training: Supervised fine‑tuning + preference/trace RL (DPO/RLAIF/RLHF), dataset curation, reward shaping, safety filters.

Tech signals

• Has shipped agents that pass real repos' test suites end‑to‑end.
• Has published papers in agentic systems/ code gen Contributions to agent frameworks or OSS evals (LangGraph/AutoGen/Guidance/LEAP, SWE‑bench variants, EvalPlus).
• Built datasets from execution traces; can show improvements from data > params.

We are committed to being an on‑site, in‑person team currently based in San Mateo