Designing trust into AI-powered workflows

Project Overview

Role

Lead UX Designer & hybrid Product Manager

Scope

6-7 months

Platform

Webapp, LLM, AI Agents

The TOA Builder is a self-service platform designed to bridge the gap between AI automation and professional accountability. Powered by Jurisage’s proprietary LLMs and trained on the Harvard Law Library database, the webapp provides a "Glass Box" user verification experience.

Responsible for:

  • Leading research synthesis and strategy pivots

  • Designing core interaction patterns and information architecture

  • Making scope and sequencing decisions

  • Partnering with Engineering closely through delivery

The Challenge

Jurisage, a Canadian AI Lawtech, sought to capture US market share. The primary obstacle was establishing trust with US litigators who are ethically bound to verify all output.

We identified Table of Authorities (TOA) creation as the strategic "beachhead"—a mandatory item that consumed hours of billable time but lacked a transparent, "practice-ready" AI solution.

The outcome

We designed a self-serve TOA generator rooted in Explainable AI (XAI) principles. By prioritizing "Proof, not promises," the UX became our most effective sales tool, converting skeptical practitioners into early adopters through radical transparency.

Success metrics:

82%

Automation Acceptance Rate (AAR)

(proving users trust the AI’s suggestions enough to keep them in the final brief.)

< 4%

Downstream Correction Rate on AI-generated citations.

82%

System Transparency Score

(via Likert surveys).

Design Decisions

01 Synchronized Duel Panel

By creating a 1:1 visual mapping between AI output and source evidence, we provided Local Explainability to eliminate the "Black Box" effect. The left panel renders the original document as a "Source of Truth," while the right panel displays the AI-generated Table of Authorities.

02 Contexual Page Jumping & Uncertainty Flagging

Rather than hiding AI uncertainty, the UI explicitly flags low-confidence matches. Clicking a flag triggers an automated "Page Jump," instantly scrolling the source document to the exact citation coordinates.

03 Transparent Processing (Narrative Loading)

I replaced static loading spinners with a Progressive Disclosure Narrative. The UI communicates by narrating the "effort" of the AI, we justify the processing time and educate the user on the rigorous multi-step verification occurring behind the scenes.

04 HITL verification

For citations requiring verification, I designed a Human-in-the-Loop (HITL) resolution suite with three distinct actions: Confirm, Edit, or Reject.

This preserves User Agency. The attorney remains the "Final Authority," while their interactions create a reinforcement learning loop that improves the model’s local accuracy for future Canadian-to-US market nuances.

Impact

91%
Reduction in end-to-end task time

(from 120 mins manual to 10 mins automated).

30%
MoM new user increases

organic growth without ads.

22% Market interest Conversion Rate

from "One-Off" TOA tool users to Gen 2 platform waitlist sign-ups.