Agentic due diligence
Multi-agent review that passes evidence between steps, checks itself and escalates to a lawyer.
Architect and lead
Read the related article →Case studies from production work and research, shown as patterns where the details are confidential: the architecture, the decisions and my part in them. Below them, open projects built on public data that you can explore directly.
Architecture and decisions, not client details.
Multi-agent review that passes evidence between steps, checks itself and escalates to a lawyer.
A semantic layer over a firm's legal knowledge, designed so agents can reason over it.
Explaining drug-repurposing predictions with filtered evidence paths through a biological knowledge graph.
Open work on public data.
A 3D map of the most-downloaded open models on Hugging Face. Each model is linked to the base model it was fine-tuned, merged or adapted from, and each base model to the lab that released it. Pick a model to trace its family, or a lab to see how far its releases have spread.
Built from the public Hugging Face Hub API, using only the base_model each repository declares. Quantised builds are counted, not drawn.
A 3D map of an open legal-industry ontology: about 18,000 concepts, from areas of law and courts to documents and parties. Each concept is joined to the one above it and to the concepts the standard relates it to. Open a category to see its structure, or search for a concept to trace its path to the top.
Data from SALI's Legal Matter Standard Specification (LMSS). Names and definitions are SALI's; the view is my own build. All data comes from open sources online, with no confidential data.