Why the Future of AI May Be Smaller: The Rise of Domain-Specific Models

Small, domain-specific AI models are coming for legal work. AI can read a contract in seconds — so why won’t an in-house lawyer trust it? Ask the same question a week later and you might get a slightly different answer. And the lawyer, not the model, carries the liability. In this episode of Talking AI, Matt Paige sits down with Emad Khazraee, co-founder and CTO of RiskVantage AI, and previously VP of AI at Xometry and a data science and AI leader at Turing. Emad spent years arguing against building a legal AI company: wrappers have no moat, and the frontier labs will ship a better one overnight. He explains the architecture that changed his mind — neuro-symbolic AI, where a deterministic legal-risk ontology does the reasoning and small domain-specific language models handle document understanding and drafting. They get into why a nine-billion-parameter model running sub-second on a commodity GPU can match a frontier model inside one domain, how subsidized token pricing is distorting the legal AI market, why RiskVantage AI sells to procurement instead of to law firms, and what happens to the billable hour when the work takes a fifth of the time. Watch or listen to the full episode: https://hatchworks.com/talking-ai/ --- Key Moments: 1:30 — What ChatGPT can’t know about your company 8:12 — Why he refused to build a legal AI company 10:22 — Subsidized tokens and the Uber comparison 16:56 — “I am on the hook for the liability” 22:13 — When you should never use an LLM 29:00 — Neuro-symbolic AI, explained 31:03 — Don’t use a missile to hit a fly 35:47 — A 9B model that matches the frontier in-domain 38:00 — Does the billable hour survive? 44:00 — The pro se surge and access to justice --- Key Links RiskVantage AI: https://riskvantage.ai/ Connect with Emad on LinkedIn: https://www.linkedin.com/in/khazraee/