About this role
Designing and hardening multi-step, tool-using agent loops forms the core of daily operations on Sourcegraph's Code Understanding team. The person in this position will tackle complex technical challenges such as model selection, evaluation design, retrieval optimization, and latency reduction to improve agentic experiences for enterprise users at scale. Operating as a staff-level technical leader, the successful candidate will set architectural standards, mentor peers, and take end-to-end ownership of the most ambiguous problems in agentic systems and production machine learning.
The Code Understanding team builds surfaces like Deep Search, Query Assist, Smart Hovers, and guided diff reviews that connect developer workflows with advanced model intelligence. Someone who thrives in a globally distributed, high-agency environment with strong customer connection will do well here. The ideal candidate brings deep expertise in balancing model costs, response latencies, and output accuracy to deliver reliable, production-ready AI products.
What you will do
- Design and harden multi-step, tool-using agent loops to transform experimental concepts into reliable, observable, and affordable enterprise-scale products.
- Establish pragmatic evaluation frameworks, targeted smoke tests, and metrics to measure answer quality without introducing unnecessary noise.
- Determine model placement, drive upgrades, and execute fine-tuning pipelines when warranted by product requirements.
- Advance retrieval, ranking, context window management, and citation mechanisms to ground models securely in customer codebases.
- Treat cost and latency as core product features by profiling, distilling, caching, and right-sizing models for sustainable operation.
- Mentor teammates, conduct code reviews, and raise the broader engineering organization's fluency in building with models.
What you need
- Demonstrated experience as a staff engineer and technical leader with hard-won skills across production machine learning, evaluation, and agent systems.
- Hands-on ownership of a complete production model lifecycle, including dataset construction, training or fine-tuning, evaluation, rollout, and monitoring.
- Fluency in building multi-step agent loops and managing retrieval pipelines, context engineering, and prompt strategies.
- Ability to operate at staff scope by solving ambiguous, high-risk problems, setting technical standards, and influencing direction beyond the immediate team.
Skills and tech stack
Questions candidates ask
What are the location and remote work policies for this role?
Sourcegraph operates as a globally distributed team. While hiring is possible almost anywhere in the world, there is a preference for candidates residing in Europe or North America, and working hours must overlap with EST for at least 20 hours per week.
What will I actually be working on regarding AI and agentic systems?
You will design and harden multi-step, tool-using agent loops, handle model selection and fine-tuning, optimize retrieval and context engineering, and manage per-user economic budgets for cost and latency across products like Deep Search.
What does the timeline look like for onboarding and making an impact?
Within the first month, you will get pipelines running locally, join the on-call rotation, and land initial improvements. By three months, you will own a meaningful agentic slice of the product and establish model change standards, and by six months, you will serve as the recognized technical authority for agent engineering on the team.