Tom Smoker is the technical founder of WhyHow.AI, where he builds knowledge-graph-powered litigation intelligence to help lawyers identify emerging cases and investigate complex claims. His systems combine structured legal knowledge, traditional machine learning and tightly controlled AI agents to make probabilistic technology more dependable in a profession demanding accuracy and accountability.
Smoker began working with graphs, ontologies and expert systems in Australia around 2016. At the University of Western Australia, he researched knowledge graph embeddings and automated reasoning, taught data science and helped develop its master’s curriculum. His 2021 research with Tim French introduced a probabilistic framework for representing uncertain beliefs and relationships. He also worked as a consulting data scientist and machine-learning engineer, including projects at BHP spanning mining and processing operations.
At WhyHow.AI, Smoker helped develop rule-based retrieval, an open-source package that constrains vector searches through explicit metadata filters. His GitHub profile also highlights Knowledge Table, a company project for extracting structured information from documents. He contributed to PatientSeek, a medical-legal model supporting reasoning over patient records.
How he makes legal AI accountable
- Purpose-built knowledge graphs: Smoker organizes products, ingredients, injuries, complaints and jurisdictions around a particular legal workflow. These compact, inspectable graphs preserve domain-specific meanings, support natural-language queries and give lawyers visual access to relationships buried in extensive document collections.
- Compounding agent error: He decomposes legal work into individually testable steps with explicit schemas, guardrails and human review. Five sequential agents operating at 95 percent accuracy produce approximately 77 percent end-to-end reliability when errors are independent—a powerful argument against treating chained agents as autonomous legal experts.
- Personalized litigation discovery: Public complaints about problems such as vehicle fires become structured signals grouped by model year, location and defect. WhyHow.AI adapts its graphs and reports to each firm’s priorities, incorporating lawyers’ feedback as potential cases develop.
- Machine learning with language-model connective tissue: Conventional models perform filtering and analysis; language models link workflow stages, interpret requests and produce lawyer-ready reports. Smoker’s approach to litigation agents keeps consequential judgment with legal professionals.
