Research engineers and research scientists are the hardest technical roles to hire in the AI/ML space. The candidate pool is genuinely small. The evaluation process is genuinely different. And the compensation requirements are genuinely high.
Most startups that need these roles underestimate all three. Here's what you actually need to know.
The titles are used inconsistently, but here's how we see them in practice:
Research Scientist: Drives the research agenda. Formulates hypotheses, designs experiments, publishes results. Strong theoretical ML background — typically a PhD from a top program or strong publication record. May write code to validate ideas but isn't primarily responsible for shipping production systems. At a startup, usually works closely with a technical co-founder who has a research background. Research Engineer (sometimes called Applied Research Engineer): Bridges research and production. Implements ideas from research papers, runs large-scale experiments, and gets models to a state where they can be productionized. Usually has strong engineering skills alongside solid ML knowledge. More common hire at startups because they can both advance the research and ship. ML Engineer (Applied): Takes outputs from research and ships them. Less focused on novel research, more focused on productionizing models reliably. Often the right next hire after the research engineer.Most Series A/B AI startups need a research engineer, not a pure research scientist — unless research output is the core product (e.g., a foundation model company).
Research talent is expensive. For companies that aren't named OpenAI or Google DeepMind, here's the current market:
| Role | Base Range | Equity (Series A) | Equity (Series B) |
|---|---|---|---|
| Research Engineer | $200K–$280K | 0.2–0.7% | 0.1–0.3% |
| Senior Research Engineer | $250K–$340K | 0.4–1.0% | 0.15–0.5% |
| Research Scientist | $220K–$300K | 0.3–0.9% | 0.15–0.4% |
| Staff Research Scientist | $290K–$400K | 0.6–1.5% | 0.25–0.7% |
These are ranges from our data across recent placements. Candidates leaving Google Research or Anthropic often have unvested RSUs in the $200K–$600K range — you need to have a real conversation about how the equity story competes.
Research hiring is different from product engineering hiring. The goal is to evaluate intellectual capability, research judgment, and the ability to execute on novel problems — not to assess whether someone can implement a binary search tree.
Round 1 — Research conversation (45–60 min). Ask them to walk you through a paper or project they're proud of. Listen for: depth of understanding (vs. surface-level summary), ability to explain trade-offs, and whether they have opinions about what they'd do differently. Researchers who can't critique their own work are rarer than you'd hope. Round 2 — Research presentation or paper discussion (60–90 min). Either: they present one of their papers/projects to your team, or you give them a relevant paper from your domain and ask them to critique it. The second is useful for evaluating candidates without strong publication records — can they engage critically with unfamiliar research? Round 3 — Technical implementation session (60 min). A focused engineering problem relevant to your stack. Not algorithms — something like "implement a simple version of X from this paper" or "debug this experiment setup." Evaluates whether research knowledge translates to working code. Round 4 (optional) — Research direction conversation. For senior hires: how do they think about your problem space? What would they work on in the first six months? What's the biggest unsolved problem in your domain? This surfaces strategic thinking beyond execution capability.Research talent at this level has real options. The pitch to a candidate from a top lab needs to be more than "good equity":
Research freedom. Larger labs have more resources but also more constraints on what research is prioritized. At a startup, a research engineer might have more direct influence on research direction. Publication continuity. For researchers, being able to publish is important to maintaining career optionality. Clarify early whether your company supports publication and on what timeline. Direct product impact. Some researchers are frustrated by how long it takes for research to translate to product at large labs. At a startup, the path from research to product is shorter. The team. Strong researchers want to work with other strong researchers. Your best recruiting asset for a research hire is the research credentials of the people already on your team.We've placed ML research engineers and applied scientists at AI-native startups across computer vision, NLP, reinforcement learning, and multimodal AI. Our sourcing goes beyond LinkedIn — we reach candidates through direct outreach, conference networks, and trusted referrals from prior placements.
Average time to hire for research engineering roles: approximately 5–7 weeks (longer than our average due to the smaller pool and longer evaluation process).
Q: How do I evaluate a research engineer without a strong publication record? A: Give them a relevant paper from your domain and ask them to critique it — not summarize it, but evaluate it. Strong research engineers can engage critically with unfamiliar work, find limitations, and suggest alternatives. The ability to do this doesn't require a CV full of publications. Q: What's the difference between a research engineer and a machine learning engineer? A: A research engineer works on novel problems where the right approach isn't known in advance — designing experiments, implementing new architectures, pushing the state of the art. An ML engineer (applied) productionizes known approaches — scaling, reliability, inference optimization. The research engineer is usually the rarer and more expensive hire. Q: How do I compete with OpenAI or Anthropic for research talent? A: You probably can't compete on base compensation at the highest levels. The pitch has to be different: research freedom, publication opportunity, direct product impact, and the size of the equity stake at an earlier stage. Some researchers prefer the smaller-team dynamics and faster feedback loops at a startup over the scale and resources of a top lab. Q: How long does it take to hire a research engineer at a startup? A: Plan for 6–10 weeks with a focused recruiting partner. The evaluation process is longer than for product engineering roles, and the candidate pool is smaller. Compressing below 6 weeks is possible but requires an unusually well-connected sourcing approach and a fast-moving interview process.For the latest engineering compensation benchmarks, levels.fyi and The Pragmatic Engineer are the most cited sources.
Related: How to Hire a Senior Backend Engineer at a Series B Startup · How to Hire a Staff Data Engineer at a Series B+ StartupTell us about your open roles and we'll start sourcing within 48 hours.
