ML & Agentic Systems Engineer [IC4]
Remote • Remote
Posted 12d ago
Job Location
Remote
Tech Stack
Remote Work Policy
Fully remote
Categories
AI Agent Engineer
About the job
Sourcegraph is building AI tools to address challenges in the software industry, especially as codebases grow and agentic development becomes prevalent. The Code Understanding team focuses on areas like Deep Search, Query Assist, and Smart Hovers, which are powered by agent engineering. This role is for a staff-level technical leader responsible for the direction of models, evaluations, and agentic systems within the Code Understanding team. The goal is to improve product performance, speed, and cost-effectiveness by integrating deterministic systems with AI. This position offers significant agency over technical direction and a direct impact on product development.
Responsibilities
- Design and harden multi-step, tool-using agent loops for current and new agentic experiences.
- Develop and implement pragmatic evaluation strategies for agentic products.
- Decide on model selection, drive model upgrades, and fine-tune models when appropriate.
- Advance retrieval and context engineering to improve model accuracy and verifiability.
- Optimize for cost and latency by profiling, distilling, caching, and right-sizing models.
- Collaborate with customers and own features end-to-end on a senior-leaning team.
- Land improvements to models, prompts, retrieval paths, or evaluations within the first month.
Requirements
- Staff-level technical leadership experience.
- Production ML and evaluation expertise.
- Experience building production agent systems.
- Ability to own complex and ambiguous problems.
- Experience setting standards and influencing technical direction.
- Proficiency in agent engineering, blending software engineering, machine learning, and statistics.
- Experience with model selection, upgrading, and training/fine-tuning.
- Experience with retrieval and context engineering for grounding models.
- Focus on cost and latency as product features.
- Ability to work with customers and own features end-to-end.