Research Engineer — Agent Architectures (Coding & Autonomous Systems)

Role Overview

Lexsi Labs is a leading frontier AI lab focused on building aligned, interpretable, and safe superintelligent systems. The mission is to build safety-aware autonomous systems in the near term, spanning AI alignment methodologies, interpretability-led system design, and foundational model research. This role focuses on researching the systems surrounding models, specifically using coding agents as a primary testbed to manage action spaces, memory schemas, and evaluation frameworks.

Responsibilities

• Design action spaces, tool surfaces, context construction policies, and control topologies.
• Develop memory schemas for execution state, compaction policies, and retrieval under closed-world constraints.
• Construct evaluation tasks from real repository history with executable verification and failure taxonomies.
• Conduct controlled experiments on architecture, including hypothesis formation, ablation studies, and establishing gains across repositories.
• Collaborate with alignment and interpretability researchers on post-training and behavioral evaluation.

Requirements

• Two or more years of experience building agents that run against real workloads.
• Deep understanding of the agentic landscape including ReAct, LangGraph, and LangChain.
• Experience building evaluation datasets or harnesses and working with SWE-bench class benchmarks.
• Proficiency in code intelligence, including ASTs, tree-sitter, and static/dataflow analysis.
• Strong backend skills with Python, sandboxing, and containerization technologies like Docker, gVisor, or Firecracker.
• Knowledge of observability tools such as OpenTelemetry and distributed tracing.
• Fluency in post-training methods like SFT, DPO, and RL, as well as distillation and quantization.

Skills

• Python
• Docker
• LangChain
• OpenTelemetry
• LLM Post-training