Staff Research Engineer – Recursive Self-Improvement Lead

Sequen · via Devremote

💰 $350,000 - $400,000/yr 🌎 Remote ⌛ Posted 5 days ago

Define the research roadmap for recursive self-improvement and decide what ships
Design and test exploration policies, compute allocation across parallel agents, and agents that revise their own instructions and strategy
Turn recorded research sessions into training and evaluation signal
Define what constitutes better research and build evaluations with the RRI team
Make knowledge learned in one session improve subsequent sessions across tasks and clients
Track RSI, autonomous research, and agentic LLM literature and turn promising ideas into measured experiments
Work with applied scientists using RRI on real client problems to identify agent shortcomings
Write code and ship successful research into systems used by clients in production
Help grow the Recursive Self-Improvement team

7+ years of experience in applied ML research or research engineering
Hands-on work in at least one of LLM agents, AutoML, meta-learning, recursive self-improvement, or a closely related area
Experience taking research ideas from prototype to a measured improvement in a shipped product or production system
Experimental rigor with baselines, ablations, seeds, and confidence intervals
Production-quality Python and ability to build experiments end to end
Understanding of frontier model behavior in long agentic loops, including context limits, compaction, tool use, failure modes, and cost
Ability to take accountability for a research direction in an early-stage team
Experience leading or mentoring a small research team
Peer-reviewed publications or widely used open-source work in agents, RL, or AutoML may be advantageous
Experience with reinforcement learning or LLM post-training, such as RLHF, reward modelling, or training agents with RL may be advantageous
Experience building evaluations or benchmarks for LLMs or agents may be advantageous
Background in search, recommendation, or learning-to-rank models may be advantageous
Experience with multi-agent systems or agent swarms may be advantageous