Post-Training Research Scientist (LLMs) — Experimental Track

Job description

**About us** Vetto is a global talent platform connecting top-tier professionals to high-impact AI projects around the world. Our mission is to build trust, quality, and long-term value in the AI ecosystem - for both exceptional talents and companies operating at the frontier of technology. **About the role** This role sits at the heart of Vetto’s mission: using high-quality human data to build AI systems that make the world better. You’ll take raw expert signals and turn it into tangible model improvement, experimenting rapidly and carving new paths in post-training. With full autonomy and no production constraints, you’ll have the freedom to try unconventional ideas and see their impact quickly. **Key Responsibilities** - Design and run **post-training experiments** on frontier and open-weight LLMs (SFT, preference-based methods, rubric-driven training) - Translate **raw annotation artifacts** (multi-step solutions, evaluations, adversarial prompts) into training-ready datasets. - Prototype **new reward signals** beyond pairwise preferences (rubrics, constraints, structured critics). - Analyze failure modes; propose **data-centric fixes** (sampling, curriculum, counterfactuals). - Build lightweight **training/eval pipelines**; iterate quickly. - Produce short internal memos: what worked, what didn’t, why. **About you** We’re looking for a researcher who thrives with autonomy, is hands-on, and brings a strong execution mindset and startup mentality. You are opinionated about data quality, pragmatic about tradeoffs, and comfortable moving quickly with incomplete information. You have strong experimental instincts — you can design, run, and interpret messy experiments and extract meaningful insights from them. **Minimum Qualification** - PhD (or equivalent experience) in ML/AI, applied math, stats, or adjacent. - Hands-on experience with LLM post-training (at least one of SFT/DPO/RLHF/RLVR). - Solid Python + PyTorch/JAX; comfortable with training infra basics. - Fluent English **Preferred Qualification** - Worked with rubric-based evaluation or tool-augmented tasks. - Experience mixing synthetic and human data. - Familiarity with failure analysis and dataset audits. **Work Model** We operate remote-first. We focus on outcomes, not where the work is done. To support flexibility and personal choice, we maintain offices in select locations as an optional resource for the team. **Location:** Flexible (EU-friendly time zones preferred) **Type:** Full-time or long-term contract **Equal Employment Opportunity** Vetto is proud to be an equal opportunity employer and values diversity at our company. We do not discriminate on the basis of race, color, religion, national origin, sex, sexual orientation, gender identity, age, disability, veteran status, or any other protected characteristic. **Type:** Full-time or long-term contract