Our research team is expanding to keep pace with a wave of frontier-facing work: internal research streams, client engagements that require real ML depth, and emerging opportunities at the cutting edge of the field. As a Research Engineer, you'll take a research direction and run with it – finding the right papers, benchmarks, and prior work, reimplementing what's relevant, and building out the process to reproduce and improve on it internally.
You'll own initiatives end to end: partnering with strategic project and technical leads to scope the work, building MVPs to validate ideas (including through human annotation and agents), and turning that work into something concrete – a customer dataset, a pilot, an internal dataset that becomes a paper or blog post, or a joint publication with a partner. You won't be handed a fully specified task list; you'll be given a direction and the autonomy to turn it into a research plan.
This is a full-time, hybrid position based in San Francisco.
What You'll Do
- Take a research direction and independently identify supporting resources – papers, benchmarks, blog posts – then implement or reimplement the relevant methods.
- Build and own the process to reproduce prior work internally and identify ways to improve on it.
- Own projects (for example, an RL/agentic environment build for a partner or a novel multimodal benchmark) end to end, including scoping, MVP implementation, and validation.
- Partner with strategic project leads and technical leads to translate ambiguous requirements into a concrete, testable research plan.
- Validate ideas through hands-on implementation, including annotating, evaluating, or sourcing data.
- Turn research directions into tangible outputs – a paid customer dataset, a customer pilot, an internal dataset, or a paper/blog post for publication or conference presentation.
- Bring an ML perspective to new opportunities — assessing technical feasibility of incoming requests and helping shape proposals where research depth is needed.
What You'll Bring
- MS or PhD in ML, CS, or a related quantitative field – or equivalent demonstrated research experience (publications, significant open-source research work, industry research).
- Real ML depth: you understand how models are trained and evaluated, not just how to call an API. You can read a paper, judge whether its claims hold, and reimplement the method.
- Hands-on experience with at least one of: RL/agentic systems, AI/ML evaluation and benchmarking, or multimodal ML.
- Strong Python and the engineering ability to build and ship your own experiments – eval harnesses, environments, infrastructure – without relying on a platform team.
- High autonomy: you can turn an ambiguous direction into a concrete research plan and notice when something's off before being told.
- Clear technical writing
Nice To Have
- Publication track record (first-author preferred).
- Experience with agent or multimodal benchmarks (OSWorld, MMMU, WebArena, SWE-bench, or similar) or building RL environments/gyms.
- Familiarity with reward modeling, reward hacking, or verifier/judge reliability.
- Familiarity with synthetic data generation or human-in-the-loop (HITL) workflows.
- Experience with cloud infrastructure and containerized environments.
- A deep RL background specifically.
Why SuperAnnotate
This is a rare opportunity to work at the intersection of frontier AI research and real production impact. You'll work on projects with frontier labs that move the needle on model performance, with your work feeding directly into the next generation of agent capabilities. You'll have the opportunity to implement projects that actually matter, publish research, and present at conferences – alongside a multidisciplinary, multinational team and collaborate with some of the most prominent labs and AI companies globally.
Skills Required
- MS or PhD in ML, CS, or related quantitative field or equivalent demonstrated research experience (publications, significant open-source research, industry research).
- Deep ML knowledge: understanding of model training and evaluation and ability to read and reimplement research papers.
- Hands-on experience with at least one: reinforcement learning/agentic systems, AI/ML evaluation and benchmarking, or multimodal ML.
- Strong Python and engineering ability to build and ship experiments, eval harnesses, environments, and infrastructure independently.
- High autonomy: translate ambiguous directions into concrete research plans and independently scope/drive projects end-to-end.
- Clear technical writing skills.
- Publication track record (first-author preferred).
- Experience with agent or multimodal benchmarks (e.g., OSWorld, MMMU, WebArena, SWE-bench) or building RL environments/gyms.
- Familiarity with reward modeling, reward hacking, or verifier/judge reliability.
- Familiarity with synthetic data generation or human-in-the-loop (HITL) workflows.
- Experience with cloud infrastructure and containerized environments.
- Deep RL background.
What We Do
The end-to-end platform to annotate, version, and manage ground truth data for your AI.
