Building AI agents that take real actions is the easy part. Building agents that get better over time - that learn from feedback, correct mistakes, and optimize toward outcomes users actually care about - is one of the hardest open problems in production AI today.
That's what this team works on. As a Staff AI/ML Engineer on our Applied Research team, you'll own the technical direction for feedback-driven learning in DigitalOcean's agentic systems: reward modeling, preference optimization, reinforcement learning, and the evaluation infrastructure needed to measure whether any of it is actually working.
This is a senior IC role with broad technical scope. You'll set direction, run experiments at scale, and close the loop between user signals and model behavior - shipping research into production, not just writing it up.
What You'll Be DoingOwn the feedback learning roadmap- Define and execute the applied research agenda for feedback-driven agentic AI - from reward modeling and preference optimization to online learning and human feedback loops.
- Translate user feedback, human evaluation data, and product signals into concrete training and optimization strategies.
- Stay close to the research frontier on RLHF, RLAIF, DPO, PPO, GRPO, and related methods and know when to apply them versus when simpler approaches win.
Build production learning systems- Design and implement learning loops that improve agent reasoning, planning, tool use, and action execution over time.
- Build evaluation frameworks that measure what matters: reasoning quality, instruction following, task success, safety, and real user outcomes - at both offline and online scale.
- Run large-scale experiments that connect model changes to measurable improvements in user experience and business impact.
Provide technical leadership- Set technical direction across modeling, experimentation strategy, evaluation design, and production readiness - without requiring direct management authority.
- Partner closely with product, engineering, design, and research teams to move work from prototype to shipped capability.
- Communicate complex AI systems clearly to both technical and non-technical stakeholders.
What You'll Add to DigitalOceanWe're looking for engineers who have shipped real learning systems - not just prototyped them. You likely bring:- 8+ years of experience building production AI/ML systems - LLMs, GenAI, agentic systems, recommendation, search, personalization, or applied research at scale.
- Hands-on experience improving AI systems through reinforcement learning, reward modeling, fine-tuning, human feedback, or preference optimization - with results you can point to.
- Strong understanding of agentic AI: reasoning, planning, tool use, action execution, instruction following, and self-correction.
- Strong software engineering in Python and at least one production systems language.
- The judgment to balance model quality, product impact, latency, reliability, cost, and maintainability - and communicate those tradeoffs clearly.
Preferred QualificationsStrong signal- Experience with agent evaluation, offline/online experiments, and human feedback loops in production.
- Direct experience with RLHF, RLAIF, DPO, PPO, GRPO, or related optimization techniques.
- Prior Staff, Senior Staff, Tech Lead, or equivalent senior IC experience.
Nice to have- Master's or PhD in CS, ML, AI, or a related field - or equivalent depth demonstrated through industry work.
- Experience with production ML infrastructure: model serving, observability, data pipelines, feature stores, or experimentation platforms.
- Research contributions via publications, patents, open-source work, or demonstrated applied research impact in RL, reward modeling, evaluation, or recommendation systems.
Compensation Range: *This is a hybrid role
#LI-Hybrid
