AI Chat Interface for Legal Professionals

I joined the project when the product was a plain AI chat embedded in Microsoft Word. Lawyers would type a question, get a wall of text back, and have no way to verify any of it. For a profession where every word carries liability, this was a non-starter.

The first thing I designed was source attribution. Every claim in the AI’s response now links back to the specific document and section it came from. Lawyers could verify without re-doing the research.

Sources tell you what was referenced, but not how the agent reached its conclusion. I designed a visible reasoning trail — the step-by-step process the agent follows, with the option to stop it mid-task or expand any step for detail.

Most AI products present every answer with equal confidence. I did the opposite. We added confidence indicators so the AI communicates its own uncertainty — because in legal work, false certainty is more dangerous than admitted doubt.\

A confidence score alone tells you ‘I’m not sure’ but not what to do about it. I designed a research checklist that appears on lower-confidence answers — actionable steps the lawyer can take to verify and increase confidence.

Up to this point, all the trust work was on the output side. Now I shifted to the input. Pre-defined prompts reduce the blank-page anxiety of a chat interface and signal that the tool understands legal workflows.

The final layer: instead of expecting lawyers to write perfect prompts, the AI asks clarifying questions with structured choices. The interaction becomes collaborative — the user works with the AI, not at it.

The next layer is control and context. I’m designing undo for AI actions, DMS file attachments so lawyers work from their own documents, and native integrations with Practical Law and LexisNexis to bring authoritative sources directly into the workflow.