callosum · London ·
Research Engineer building benchmarks and evaluation suites for agentic AI systems—designing rigorous measures for multi-turn, tool-using LLMs, turning traces into structured evidence, and building evaluation/observability infrastructure the whole company relies on.
Anthropic · San Francisco, CA | New York City, NY
jobleads · Remote DE
Emergent · IN
DRW · Montreal
About Us
We’re living through a Cambrian explosion of intelligence: new models and new chips, each specialised for different tasks, are arriving all at once. The result is a new era for AI, one of radical heterogeneity.
Callosum is the Intelligent Systems Company. We believe the next generation of AI won't be defined by any single model or chip, but by intelligent systems in which hardware and intelligence co-evolve. We are building the infrastructure that unifies heterogeneous compute across the full stack. This opens a new axis of scaling intelligence: a dynamic system that tailors itself to what each workload actually needs, whether that's speed, cost, precision, or whatever unit comes next.
The last era scaled on a different bet: one bigger model, more of the same chip, more data. That bet is running into structural limits. Frontier models offer extraordinary capability at unsustainable cost, one that today's monolithic infrastructure was never designed to serve.
Our founding principle is that intelligence comes from many specialised systems working together, not from any single component. We build the software orchestration layer that co-evolves models, workflows and silicon into one system, delivering inference tailored to every workload, and demonstrating orders-of-magnitude leaps in capability and cost.
Because our software spans the full stack, our engineering team works directly with heterogeneous accelerators and frontier silicon, including Cerebras, d-Matrix, Intel, NVIDIA, AMD, Normal Computing, Tenstorrent, GreatSky, and Mixx. We are not stopping at today's chips: each new generation of silicon unlocks algorithms that couldn't run before, and we intend to be first to them, every time. If we get it right, it will belong to everyone building on it - not to any single vendor.
In our latest funding round, we raised $100M, led by Atomico with participation from Plural, DCVC and the UK Sovereign AI Fund’s first investment. With this, we are building the infrastructure for the next era of intelligence.
We are engineers and scientists based in London, working across the full depth of the stack. We are curious, intellectually honest, and building what doesn't exist yet. If you thrive on uncharted territory and are energised by the scale of the challenge, we'd love to hear from you.
About the Role
Callosum believes that orders of magnitude improvements in AI systems will come through application-aware orchestration across heterogeneous models and hardware. Our team is developing the science that makes that possible: a principled, evidence-based method for building agentic systems automatically at unprecedented scale.
The team tackles these problems on two fronts. We build the tools to analyse and evaluate agentic systems rigorously enough to say where and why they go wrong, and we use what we learn to design better ones. We are not simply building a better harness; the best harness will change with every new model, task and generation of silicon. We are building the layer beneath it, so that design decisions follow from evidence rather than taste.
This role owns the first of those fronts. Agentic evaluation today is not up to the job: benchmarks saturate, scores move for reasons unrelated to capability, results fail to reproduce across runs, and a single number tells you nothing about which step went wrong. Every design decision the rest of the team makes rests on the quality of that instrument. You will build measures whose construct we can defend, whose variance we understand, and which localise a failure rather than merely scoring it.
The mandate is broad, and most of the questions inside it are still unanswered. You will have wide latitude to choose problems, and your results will shape what the company builds in the future.
What You'll Build
Design benchmarks and evaluation suites for agentic behaviour – multi-turn, long-horizon, tool-using, operating in non-stationary environments – including the properties that are hard to score, such as reliability and recovery from error
Build the methodology as well as the harness: statistical power, variance across runs, contamination and construct validity
Be the independent voice on the quality of our own results. Red-team our evaluations, design the sanity checks that catch a flattering number before it reaches a decision.
Turn raw traces into structured evidence. Failure taxonomies, behavioural signatures, and attribution of an outcome to the decision that caused it
Move from measurement to prediction: infer what a system can do from partial evidence, and estimate performance on a task before running it. This rests on defining what each model is genuinely good at, what benchmarks really measure, and what capability profile a real compound task demands
Build evaluation and observability infrastructure as durable instruments the whole company works from
What You'll Bring
Evidence that you can run research of your own: you take an open question, design the experiments that settle it, and produce results other people can build on
Deep hands-on experience with LLMs in agentic settings: multi-step tasks, tool use, long horizons, and the specific ways all of it breaks
Statistical discipline. Hypotheses stated in advance, uncertainty quantified, and running comparisons that isolate one variable
Strong engineering. You write the code that runs your experiments and you are comfortable working inside a large shared codebase
Strong communication skills. You are able to turn research findings into clear, prioritised guidance for the teams who will act on them, and to write them up for a wider audience
What Sets You Apart
Evaluations or benchmarks you built that other people went on to use
Broad training in empirical method, potentially from a field outside machine learning - experimental design, Bayesian inference, model selection, significance testing, uncertainty quantification
Experience with automated grading and model-based judging, and a rigorous account of its failure modes and its agreement with human raters
A published track record in a relevant field - first-author work at venues such as NeurIPS, ICML, ICLR or ACL, including the datasets and benchmarks tracks
What We Offer
Competitive Salary, determined by skills and experience
Equity & Ownership
Private healthcare
We offer Visa sponsorship and relocation benefits to hire the best in the world
We work in person at our London office. You'll have the tools, space and setup to do your best work, and if you have specific needs, just tell us
We're committed to building an inclusive workplace where everyone feels welcome, and believe in equal opportunities for all.
Zaka Vc · San Francisco, CA, USA