Building and evaluating AI systems

Case Genie, built by 67 Bricks for ICLR, lets users upload their own legal documents — skeleton arguments, for instance — and uses NLP to identify the legal concepts inside, then suggests relevant cases from the ICLR database,…

ClockHive is a hackathon-built experiment: a preprint server whose target users are AI agents, not humans. The website has no buttons and no human login — everything happens through the API, so autonomous ‘open claw’ agents (which run…

Back in 2015, Elsevier Labs adopted Databricks to build NLP algorithms against their content archive. Shared notebooks replaced fragmented individual workflows, letting multiple teams process terabytes of text and petabytes of supplementary data.

Adam Hyde built two scholarly publishing platforms largely by “vibe coding” with Anthropic’s Claude: EasyJournal, a conventional system in the mould of OJS and Janeway, and Kotahi, a collaborative-first alternative with non-linear, communication-driven workflows.

The BMJ and the AWS Generative AI Innovation Center built an editorial assistant that helps journal editors screen submitted manuscripts — supporting decisions about which papers to send for peer review, which to reject, and why. Written up…

eLife Pathways is eLife’s open-publishing-infrastructure initiative (backed by a £2.4m Wellcome investment), building on its publish-review-curate model toward modular publishing — separating a paper into parts so both humans and machines can link between them, with different composed…

JAIGP is a multidisciplinary, open-access journal built by César Hidalgo in collaboration with Claude Sonnet 4.5, with an inverted editorial policy: AI systems are the authors and humans are the prompters. It is positioned as a home for…

Paper2Agent converts computational research papers into chat agents you can interrogate — “apply your method to my data”. Works only with well-structured public code repositories; conversion takes 30 minutes to 3+ hours and about $15 per paper.

pure.science, led by CEO Adam Hyde (of Book Sprints / Coko) and CCO Laura Harvey, offers no-code workflows for publishing and research: orchestrating systems, automating decisions and deploying AI agents for accessibility checks, citation validation, research discovery and…

Silverchair and OpenSource Connections co-designed an evaluation pipeline for benchmarking Retrieval-Augmented Generation (RAG) architectures on real publisher content. Built with Python, LlamaIndex, RAGAS and LangSmith, it generates question-context-answer triplets, runs RAG models, and scores outputs on faithfulness, answer…