Research Engineer, LangSmith Engine

About Us

At Lang Chain, our mission is to make intelligent agents ubiquitous. We build the foundation for agent engineering in the real world, helping developers move from prototypes to production-ready AI agents that teams can rely on. We began as widely adopted open-source tools and have grown to also offer a platform for building, evaluating, deploying, and operating agents at scale.

With $125M raised at Series B from IVP, Sequoia, Benchmark, CapitalG, and Sapphire Ventures, we’re at a stage where we’re continuing to develop new products, growth is accelerating, and all team members have meaningful impact on what we build and how we work together. Lang Chain is a place where your contributions can shape how this technology shows up in the real world.

Today, our platform includes Lang Smith (Observability, Evaluation, Deployment, Fleet, and Sandboxes), our open source frameworks (Lang Chain, Lang Graph, and Deep Agents), and the newly launched Lang Smith Engine for autonomous agent improvement. We have 100M+ monthly open source downloads, 6,000+ active Lang Smith customers, and 5 of the Fortune 10 use Lang Smith in production (+ 35% of the Fortune 500 overall), including teams at Klarna, Clay, Coinbase, Workday, Lyft, Cloudflare, Harvey, Rippling, Vanta, Linked In, , Nvidia, and Bridgewater.

The Lang Smith Engine team is building a proactive agent engineer that analyzes production traces, identifies important failures, recommends and writes fixes, and helps prevent those issues from coming back. We’re building agents that can understand complex software systems and continuously improve the quality of other AI agents.

We’re looking for an experienced research engineer to help make the Engine agent more capable and more efficient.

You’ll study real agent failures, build benchmarks that capture what matters, run experiments to improve performance, and turn successful ideas into production. This may include prompting and agent-harness improvements, model selection, fine-tuning and post-training custom models. The focus is on measurable improvements to the overall agent.

This role also requires a understanding of production engineering and system-level tradeoffs
. Engine is a production system, so improving an agent is not just about maximizing benchmark performance—it also means understanding the impact on cost, latency, reliability, and scalability
. You’ll work in the same team with production engineers to design, test, and ship improvements that work reliably in real-world environments.

Location: SF and NYC

• Build and maintain benchmarks and evaluations that measure the quality and efficiency of Engine agents on real-world tasks.

• Design and run experiments to improve agent performance across models, prompting, context, tools, orchestration, and agent strategies
  .

• Explore and implement post-training and fine-tuning techniques when they can meaningfully improve agent capabilities, quality, or cost.

• Turn successful experiments into production improvements
  , working closely with engineers and researchers to measure impact and prevent regressions.

• Help define the ML roadmap and technical direction for improving Engine agents, and mentor other engineers through strong technical leadership.

• 4+ years of experience in ML/AI research, or a closely related field.

• Master’s or Ph.D. in a relevant scientific field.

• Hands-on experience working with LLMs and AI agents
  , including analyzing model behavior and improving real-world performance

• Strong experience designing benchmarks, evaluations, and experiments for AI/ML systems; you know how to tell whether a change actually made an agent better.

• Strong software engineering skills, with a track record of taking ideas from research prototype to measurable production impact
  .

• You have maximum agency and strong research judgment
  : you can identify high-impact problems, work through ambiguity, move quickly, and communicate your findings clearly.

• Ph.D. in Machine Learning, Computer Science or Physics.

• Hands on experience with LLM-as-a-judge, automated graders, synthetic data generation, or human evaluation
  .

• Hands…