Bentham Research is an applied research lab at the intersection of AI and the humanities. We transform expert inquiry into rigorous evaluation environments and post-training data for frontier AI.

Our scholars, researchers and engineers are building the infrastructure required to evaluate and improve how AI systems reason across philosophy, history, ethics, political thought, theology, law and global cultural traditions. We capture not only scholars’ final answers, but the deeper process of humanistic inquiry: interpretation, disagreement, revision, peer review and judgment under uncertainty.

We are looking for a product-minded Software Engineer to build the platform through which this work happens. It will enable scholars to design adversarial evaluation tasks, construct reasoning rubrics, assess model outputs, conduct blind peer review, resolve disagreements and turn expert judgment into structured datasets for evaluation and post-training.

The ideal candidate will be excited by ambitious tools for thought, excellent user experience and the challenge of translating complex scholarly workflows into software that feels clear, fast and intuitive.

• Build the core platform used by humanities scholars to create, review and refine AI evaluation tasks.
• Design workflows for benchmark construction, reference answers, reasoning rubrics, model evaluation, peer review, adjudication and quality assurance.
• Develop interfaces for complex structured information, including prompts, answers, rubrics, citations, reviewer feedback, model outputs, metadata and customer deliverables.
• Integrate frontier AI models to support task development, answer analysis, similarity checking, quality control and adversarial testing.
• Work directly with scholars to understand how expert inquiry actually unfolds—and design tools that strengthen rather than oversimplify it.
• Build systems for blind peer review, inter-rater reliability, version control, provenance and multi-scholar disagreement.
• Partner with Bentham’s research and operations teams to turn delivery bottlenecks into scalable platform capabilities.
• Help transform bespoke research workflows into a secure, repeatable and high-throughput evaluation pipeline.
• Own features end to end, from product discovery and interface design through implementation, deployment and iteration.
• Contribute to the technical foundations of Bentham’s contributor portal, evaluation engine and customer-facing delivery infrastructure.

• Strong full-stack or product engineering ability: you can understand a real user workflow, design an effective solution and ship it.
• Excellent judgment in interface design, information architecture and user experience.
• Experience building sophisticated web applications, internal tools, collaborative software, research platforms or evaluation systems.
• Strong frontend capability, ideally using React or Next.js, with sufficient backend experience to own features across the stack.
• Experience integrating LLMs into production software, including prompting, structured outputs, model selection, evaluation, review loops and failure handling.
• Familiarity with Python, APIs and structured data systems. Experience with FastAPI, PostgreSQL, pgvector or comparable technologies would be useful.
• A thoughtful approach to data provenance, permissions, auditability and quality assurance.
• High attention to detail and a preference for simple, elegant and reliable systems.
• The ability to work effectively in ambiguity and turn intellectually complex processes into coherent product experiences.
• An interest in AI evaluation and humanistic reasoning. Formal humanities training is welcome but not required.

• You have built tools for knowledge work, research, writing, expert review or collaborative judgment.
• You care deeply about making complex intellectual work feel fluid and well structured.
• You have strong opinions about what good software looks and feels like—and can turn those opinions into working products.
• You are interested in “tools for thought” and how they can support reasoning, interpretation and judgment.
• You want to work directly with users, ship quickly and see your work affect the quality of real evaluation and post-training projects.
• You have worked on an AI evaluation product, an LLM workflow platform, an expert-facing data product or sophisticated internal tooling.
• You are motivated by the opportunity to help AI move beyond reproducing human knowledge towards reasoning with greater depth, creativity and intellectual honesty.

The humanities present some of the hardest and most consequential challenges for advanced AI. Evidence is incomplete. Interpretations conflict. Context matters. Intellectual honesty may require acknowledging uncertainty rather than producing a confident answer.

Bentham brings together exceptional scholars and engineers to turn these forms of reasoning into rigorous evaluation environments and training data. You will help build the platform that makes this possible—and the infrastructure through which human expertise can shape the next generation of machine intelligence.

Please send your CV or résumé, together with a few lines about who you are, some of your favourite things you have built and why you are interested in joining Bentham Research.

Bentham Research is an applied research lab at the intersection of AI and the humanities. We transform expert inquiry into rigorous evaluation environments and post-training data for frontier AI.

We are building a global community of PhD-qualified scholars to help frontier AI systems reason more deeply across philosophy, history, ethics, theology, political thought, law, literature and global cultural traditions.

This is not annotation or content-labelling work. Bentham scholars apply the same intellectual methods developed through doctoral research: framing difficult questions, interpreting evidence, distinguishing fact from argument, engaging competing schools of thought, identifying hidden assumptions and judging when the evidence does not permit certainty.

The difference is the form of the output. Scholarly expertise is translated into structured evaluation tasks, reasoning rubrics, reference responses, adversarial challenges and post-training data that can be used directly by frontier AI laboratories.

• Pose your own substantive research questions to frontier AI models in Bentham Studio, a scholarly conversation and annotation workspace, rather than working from prewritten benchmark tasks.
• Read model answers closely, select passages and record what you notice and the reasoning behind your assessment—criticising, correcting, explaining or recognising strengths.
• Identify omissions: what a strong answer in your field should have addressed but did not.
• Review Bentham Cards—concise AI interpretations of your comments, each with a label, principle and demonstration—and correct or refine them until they faithfully represent your judgment.
• Validate the Cards you endorse, recording your confidence that another qualified scholar would agree and, where relevant, the severity or importance of the finding.
• Preserve qualifications, scope and legitimate disagreement, so that contested interpretations are represented rather than flattened into a single answer.
• Continue the dialogue by asking the model to respond to your comments or by pursuing new lines of inquiry, testing how its reasoning holds up under expert challenge.
• Ground conversations in the literature you know best by supplying primary and secondary sources and building folder bibliographies.
• Act as an independent reviewer of other scholars’ prompts and responses, assessing a fixed model answer without seeing the original scholar’s annotations.
• Where research access is granted, use Bentham Studio for your own inquiry, working with your own documents in a personal research workspace.

Bentham’s methodology is designed as an extension of scholars’ existing training, rather than a departure from it.

The work draws directly on familiar academic practices:

• Formulating consequential and intellectually demanding research questions.
• Conducting close analysis of primary and secondary sources.
• Distinguishing evidence from interpretation.
• Testing arguments against counterarguments and competing traditions.
• Making assessment criteria explicit.
• Reviewing the work of scholarly peers.
• Revising conclusions in response to criticism or new evidence.
• Acknowledging uncertainty, ambiguity and the legitimate limits of knowledge.

Bentham makes this process legible to AI systems. Final answers are only the beginning: the reasoning, revisions, disagreements and judgments that produce them can also become valuable signals for evaluation and post-training.

Peer review is central to Bentham’s approach and one of the foundations of our quality standard.

Submissions are reviewed by independent scholars for disciplinary accuracy, intellectual rigour, evidential support, fairness, ideological neutrality and the quality of the accompanying rubric. Reviewer disagreement is measured and, where necessary, referred for senior scholarly adjudication.

Items may then be tested against frontier models and subjected to adversarial review to determine whether they expose the intended reasoning capabilities and failure modes.

Scholars therefore participate both as authors and reviewers. The aim is not to force consensus where genuine disagreement exists, but to represent contested interpretations accurately and to establish transparent standards by which model reasoning can be assessed.

• A completed PhD in a humanities discipline or a closely related field. A PhD is required.
• Deep subject-matter expertise demonstrated through doctoral research, publications, teaching or an equivalent scholarly record.
• The ability to formulate difficult questions that test reasoning rather than simple factual recall.
• Strong command of evidence, sources and the major interpretive debates within your field.
• The ability to explain what distinguishes an exceptional answer from one that is merely plausible or fluent.
• Experience giving and receiving rigorous academic peer review.
• Intellectual openness and the ability to represent competing scholarly traditions fairly.
• Careful judgment when working with ambiguity, incomplete evidence and legitimate disagreement.
• Excellent written English and the ability to express complex ideas clearly and precisely.
• Interest in AI evaluation, post-training and the future of humanistic knowledge. Previous technical or AI experience is welcome but not required.

• Applied ethics and moral philosophy
• Philosophy and intellectual history
• History and historiography
• Political theory and political philosophy
• Theology and religious studies
• Law, jurisprudence and legal philosophy
• Literature, rhetoric and cultural criticism
• Anthropology and cross-cultural studies
• Classics and ancient civilisations
• Art history, aesthetics and visual culture
• Non-Western intellectual and philosophical traditions
• Interdisciplinary fields that connect the humanities with science, technology, medicine, business or public policy

• You believe advanced AI should engage with the full depth of human knowledge, not merely reproduce familiar summaries.
• You are interested in how expert judgment can be captured without reducing scholarship to simplistic right-or-wrong answers.
• You can identify subtle but consequential failures in apparently convincing arguments.
• You are comfortable saying that the evidence is inconclusive or that multiple positions remain defensible.
• You value intellectual honesty over rhetorical certainty.
• You enjoy constructive peer review and regard criticism as part of producing stronger work.
• You want your scholarly expertise to influence how the next generation of AI systems is evaluated and trained.
• You are excited by meaningful interdisciplinary collaboration with other scholars, researchers and engineers.

AI systems increasingly influence decisions involving values, institutions, history, culture and human behaviour. Yet the infrastructure used to train and evaluate those systems remains weighted towards problems with answers that can be verified automatically.

The humanities address a different class of problem: evidence can be incomplete, concepts change across cultures and periods, interpretations conflict, and good judgment depends upon context.

Bentham places scholars at the centre of solving this problem. Our ambition is to ensure that the world’s intellectual traditions—and the rigorous methods developed to interpret them—help shape the future of machine intelligence.

Please send your CV, a brief description of your doctoral research and a few lines explaining why you are interested in contributing to Bentham Research.

Where possible, please also include links to relevant publications or examples of your scholarly work.