Founding Research Engineer
Axiōma Search · London Area, United Kingdom
Apply & track with Apply EdgeFounding Research EngineerDeep Tech Startup | LondonCompensation: up to £120k base + founding equity packageAboutBuilding a chip takes hundreds of engineers and years of work. Physical design — the placement, routing, and timing of transistors — is one of the hardest stages: thousands of interdependent decisions, sparse domain data, and no general-purpose AI that understands circuit structure. This role exists to change that.This is an early-stage startup backed by top EU investors, building an AI reasoning system for physical chip design. You'd be the first AI hire, working directly alongside the co-founding team and a founding chip design architect. What you'll doDesign and build AI reasoning systems that understand circuit structure and make design decisions autonomouslyBuild post-training pipelines — fine-tuning, reward modelling, and feedback-driven learning on proprietary chip design dataDevelop synthetic data generation strategies for a domain with limited open-source examplesDrive data strategy, model architecture, and inference optimisation across the full AI systemWork directly with the founding chip design architect to understand problem structure and translate it into training signalsShape technical direction alongside the founders — framing problems, proposing approaches, designing experimentsWhat you'll needDeep, hands-on experience building AI systems for hard, novel problem domains — reasoning, mathematics, biology, physics, or similarAbility to build reasoning models from scratch: post-training, RL, fine-tuning, data strategy — not just deploying existing pipelinesStrong fundamentals across model architecture, training dynamics, and inference optimisationFirst-principles mindset: given a new structured problem, you can understand it and build an AI system to solve itStartup mindset: comfortable with ambiguity, able to move fast without a large team behind youOptional BonusExperience with symbolic reasoning, graph-structured data, or formal methodsBackground in post-training for reasoning or maths-style modelsSynthetic data generation for low-data domainsShortlisted candidates will be contacted within 48 hours.