We’re building the platform that makes the $50b healthcare AI market safe, reliable and transparent by continually productising frontier LLM capability and interpretability research At Parsed, we're supercharging, interpreting and ‘robustifying’ LLMs for every specific use case in healthcare. We turn black-box models into transparent systems using frontier mech interp research so that clinicians can trust, regulators can approve, and patients can rely on such models. Just like Stripe secured and simplified payments, Parsed is setting the new standard for safe, robust and useful AI in healthcare. Parsed is building the primitive required for the fully autonomous AI clinician. LLMs should be used for more than just admin in medicine, we are building the unlock to make this a reality. At Parsed, we are supercharging LLMs that don’t just predict, but models that reason, plan, and operate with full transparency. By surgically inspecting and steering LLMs internal computations, we align models to think like real clinicians. Our platform gives developers complete visibility and control over AI decisions, making us the foundation for the inevitable AI clinician. What we do We are looking for a Founding Research Engineer to join our team of ML researchers, Rhodes Scholars, PhD candidates, neuroscientists and medical doctors to help develop our healthcare LLM platform which: Supercharges open-source LLMs for specific tasks through data-efficient alignment of evaluation models and discrete optimisation Interprets internal reasoning using mechanistic interpretability techniques such as sparse autoencoders and probes Robustification of outputs through ongoing adversarial testing and model steering As a deeply academic team, we of course value general purpose AI research. However, at Parsed, we are in the business of actually saving lives by unlocking the latent ability of LLMs do be useful in healthcare. Your daily contributions/pushes could directly save and improve patient lives - there aren’t many ML positions where the link between code and impact is so tangible. Responsibilities Design and train task-specific reward and evaluation models using minimal supervision and working with our medical knowledge team to incorporate domain-specific priors (e.g. clinical guidelines, structured medical knowledge) Develop and optimise sparse autoencoder architectures for identifying and manipulating latent circuits in large-scale language models Implement and extend probing techniques (linear, nonlinear, causal) to audit and interpret internal model representations related to clinical reasoning, safety, and bias Engineer closed-loop systems for adversarial testing and behavioural feedback —automatically stress-testing models with distributionally shifted, ambiguous, or adversarial clinical prompts Build pipelines for discrete optimisation of LLM behaviour — e.g., prompt architecture search, decision routing, or tool-use strategies via bandits, evolutionary search, or Bayesian methods Integrate steering and editing mechanisms (e.g. concept erasure, activation patching, editing-by-intervention) to fix or improve model behaviour with high precision Develop internal frameworks for interpretable evaluation harnesses to measure consistency, faithfulness, and safety across tasks like summarisation, triage, and recommendation You won’t just build, you’ll shape the core R&D culture of Parsed. You’ll have a defining voice in architecture, research direction and how we deliver AI-driven healthcare. We encourage open-ended interpretability research, with the aim of submitting to a top-tier conferences. Qualifications Convinced that the most scalable and impactful way to change healthcare is by unlocking LLMs latent ability to perform complex clinical workflows First-author publications in premier ML venues e.g. NeurIPS, ICML, ICLR etc. Knowing when to get to “good enough” (speed and experimentation) and knowing when it needs to be “perfect” (sequenced execution for mission-critical systems) Some production engineering experience Ability to proactively theorise and get to experimentation within hours Logistics Location policy: In London. If you are exceptional, we will consider remote. Visa: If you are exceptional, we will sponsor you. Benefits Truly top-of-market early-stage equity Competitive salary ♀️ Excessive health allowance (we are a health company after all) Catered lunches and a stocked kitchen Monthly book allowance Healthy commute stipend support (if you bike/walk to work - we’re a health company and we want to support) Laptop tools you need to succeed Learning & development budget ️ Team-building events Parsed is the most scalable way to actually improve patient lives. Our core team comprises of a dual-trained clinician and PhD ML researcher, a mechanistic interpretability researcher specialising in LLM circuit tracing, and a computational neuroscientist PhD researcher focused on interp for natural intelligence. The future of medicine is clinical AI that can think, explain, and act safely. Deep dual expertise in both medicine and interpretability is essential for this mission. This is the DNA of our founding team. Whilst building for enterprise at our previous startup, and founding/growing charities, we’ve turned down Oxford PhDs, and rejected offers to work at frontier AI labs. We’ve never second guessed these decisions, because we know there is not better way to transform healthcare than by building Parsed. We’re backed by the best. Our lead investor LocalGlobe is the most successful ‘unicorn backer’ in UK/Europe. Notable angels include the ex-director of DeepMind, co-founder of HuggingFace, ex-chair of the NHS. We are building towards a future where medical models operate at human-level intelligence with machine-level scale. We know this is only possible with the world class talent — that’s why we are assembling a lean team who are doing their life’s work. Our life’s work being delivering the safest, most impactful, and most scalable path to better patient outcomes.