AI Researcher, Ivo

• A Ph.D. in Computer Science, Engineering, Mathematics, Physics, or a related quantitative field — or equivalent industry research experience with a comparable track record.
• Evidence of exceptional ability — a paper, shipped system, open-source contribution, competition result, or hard problem you cracked that puts you meaningfully ahead of your peers.
• Deep, hands-on experience in deep learning research and development, particularly with LLMs. Strong working knowledge of modern frameworks (PyTorch, JAX, or TensorFlow) and the surrounding open-source ecosystem.
• Expertise in at least one of: agentic systems, reasoning, parameter-efficient fine-tuning (PEFT) methods, quantization, inference optimization (e.g., speculative decoding), hallucination mitigation, novel architectures in deep learning, or robust evaluation methodology for LLMs.
• Excellent communication skills, with the ability to articulate complex research findings clearly to both technical and non-technical audiences.
• A bias toward action: you ship rather than perfect, you measure rather than guess, and you'd rather have a working prototype today than a polished plan next week.

Desirable

• You'll own a research roadmap end-to-end: identifying the right problems, designing experiments, prototyping, and shipping the winners into production alongside the engineering team
• Concretely, that looks like:
• Advance the core AI platform
• Design and implement novel approaches to the problems at the heart of Ivo's product: reasoning over long-context legal corpora, contract comparison and redlining, information extraction, and automated drafting and editing
• Make our models trustworthy
• Conduct research on procedural hallucination detection and resolution, calibration, and explainability
• In a domain where a single fabricated citation can sink a deal, the bar for groundedness is uncompromising — your job is to keep raising it
• Push frontier techniques into production
• Explore and apply advanced fine-tuning, PEFT, and distillation techniques to make our models faster, cheaper, and more accurate on legal- specific tasks
• Evaluate emerging work in agentic systems, long-context modeling, and reasoning, and figure out which ideas actually move the needle for our customers
• Build the evaluation infrastructure
• Design and maintain datasets, benchmarks, and evals for training and measuring model performance on complex legal text
• Define the metrics that matter, and hold the team to them
• Partner closely with Engineering and Product to take prototypes from notebook to production, write internal reports that influence the technical direction of the platform, and present findings to both technical and non-technical audiences across the company