William Boan
Lawyer → Engineer → Founder → Frontier AI
I enter hard, ambiguous domains, learn what must be learned, build what must be built, and iterate until reality moves.
As co-founder and CTO of Contend, I built a legal AI assistant for people who couldn’t afford a lawyer – along with the multi-stage architecture and the evaluation framework that made its answers trustworthy.
Now I’m focused on frontier AI: On advancing model reasoning, reliability, and trustworthiness – to benefit humanity.
Selected Work
-
Contract-Review Agent
ProjectAn agentic system that reviews .docx business contracts through an Anthropic tool-use loop – built around a deterministic parser whose typed-block view is the agent’s view of the document.
-
Contend Project Write-Up 1
Contend Chatbot
Deep diveHow I architected Contend’s core product, a legal chatbot, to reliably guide non-expert users through a structured consultation – a two-level LangGraph state machine driven by dynamically generated conversation plans.
-
Contend Project Write-Up 2
Contend Evaluation Framework
Deep diveHow I built – and validated – the evaluation framework behind Contend’s legal chatbot, including a meta-evaluation layer that measures the evaluators themselves.
-
Contend Project Write-Up 3
Contend Content Generation System
Deep diveThe knowledge-grounded content system I built at Contend – a ~2,000-topic legal taxonomy plus an 11-phase generation pipeline producing a coherent, accurate content ecosystem at scale.
-
Robust Reasoning
Research memoWorking notes on why current models stay brittle at robust reasoning – a substrate-conversion framework for the costs of moving between sub-symbolic, natural-language, and formal representations.
-
A Transformer From Scratch
ProjectA decoder-only transformer implemented from scratch in PyTorch, wrapped in a production-grade training harness – Hydra configs, a cosine LR schedule, checkpointing, W&B logging, tests, and CI.