JT Assistant: A Smarter AI Chatbot for Legal Workflows
Helping prosecutors and court staff navigate complex systems faster and with less friction
Overview
JT Assistant is an AI-powered chatbot embedded within a legal case management system used by courts and prosecutors. It provides real-time, context-aware support that helps users find information, complete tasks, and move through complex workflows without leaving the case they are working in.
The focus was not simply introducing AI into the product but on reducing the effort required to complete everyday tasks in an environment where speed, accuracy, and focus are critical.
Complex systems slow down critical work
Legal professionals operate in fast-paced, high-pressure environments, yet the systems they rely on often introduce friction at every step. Tasks that should take seconds, like locating a document or confirming case details, often require navigating multiple screens or running manual searches.
Over time, these small inefficiencies compound and slow down critical work. The issue was not a lack of functionality, but the amount of effort required to access it.
Reframing the Opportunity
Instead of approaching this as a chatbot feature, I reframed the problem around workflow efficiency.
How might we bring the right information and actions directly to users at the moment they need them
This shifted the direction toward a more integrated assistant that supports users within their existing workflow, rather than pulling them away from it.
My Role
I led the end-to-end design of the JT Assistant experience from early discovery through final UI and validation.
I partnered closely with Product and Engineering to define the experience, align on constraints, and make tradeoffs that balanced usability with the realities of a complex legal system.
Understanding User Needs
To identify where AI could provide meaningful value, I focused on how users actually worked in real scenarios rather than how the system was designed to be used.
Stakeholder Workshops
Facilitated working sessions with court staff to understand internal processes, dependencies, and operational constraints. These conversations helped uncover where delays were happening and where an assistant could realistically reduce effort.
User Interviews
Conducted one-on-one interviews with prosecutors to understand their day-to-day workflows. Many described relying on memory, repetition, and manual navigation to complete routine tasks, which highlighted clear opportunities to reduce friction.
Mapping workflows
Mapped end-to-end workflows to visualize how users moved through the system. This made it easier to identify where time was being lost and where interruptions were most common.
Affinity Mapping Results
Organized research findings into key themes to identify patterns across users. This helped clarify which problems were consistent and worth prioritizing versus isolated edge cases.
What We Discovered
Users were not struggling with what to do, they were struggling with how long it took to do it.
A key insight was that users were not asking for a new tool. They were looking for a faster way to complete the tasks they already knew how to do. This reinforced the need for a solution that fits naturally into their workflow.
Low Fidelity
Key Focus Areas:
Establish basic layout with sidebar navigation and main content area
Define primary content blocks and their relative importance
Identify key user interaction points
Design Decisions
Worked closely with Product and Engineering to shape decisions based on both user needs and technical feasibility.
This included balancing the visibility of the assistant with available screen space and ensuring the experience could evolve as the AI became more capable.
Designing for Context
Designed the assistant to automatically reference the active case, so users do not need to restate information or perform additional searches.
This small shift significantly reduced repeated effort and made interactions feel more seamless.
Early Prototypes
Keeping Users in Their Workflow
Placed the assistant in a persistent side panel so users could interact with it while continuing their work.
This reduced context switching and allowed the assistant to feel like part of the system rather than a separate tool.
Reducing Friction with Suggested Actions
Introduced suggested prompts to help users get started quickly.
Instead of encountering an empty input field, users are guided toward common actions based on their current context, reducing effort and increasing adoption.
Onboarding & Empty States
Designed onboarding to clearly communicate what the assistant can do and when it is useful.
Empty states were used as an opportunity to guide users toward meaningful first actions rather than leaving them unsure where to begin.
Intelligent Search
Enabled natural language search so users can find documents and case details without needing exact keywords or navigation paths.
Results are structured to support quick scanning and immediate action.
Natural Conversation Flow
The chat interface feels natural while providing structured data and actionable suggestions.
Full conversation view with contextual responses and suggested follow-up actions.
Design Evolution
Explored multiple approaches to determine the right balance between visibility, usability, and integration within the system.
Digital Wireframe Concepts
Exploring different layout approaches and information hierarchies. Each concept prioritizes different aspects of the user experience to evaluate which approach best serves our users' needs.
Iteration 1: Floating Chatbot – A common, low-friction entry point that's easily accessible but can be easily missed.
Iteration 2: Persistent Panel – A dedicated sidebar that increases visibility and space for complex interactions, while remaining a generic chat experience.
Iteration 3: Context-Aware Assistant – The final, most integrated solution. The AI is proactive, suggesting specific actions (like finding missing signatures) based on the case's current context.
High-Fidelity Design Exploration
Refined the interface to improve clarity, consistency, and usability while ensuring it aligns with the broader system.
Complete Product Ecosystem
Integrated the assistant into the overall product experience so it feels like a natural extension of the system rather than an added feature.
Complete Product Ecosystem
Validating with Real Users
What We Tested
Finding case information
Uploading documents
Navigating workflows
Moderated Testing
Observing user interactions
What We Learned
Suggested prompts improved usability
Context awareness increased trust
Users expected action-based capabilities
Feedback Collection
Gathering insights in real-time
What Changed
Refined prompts
Improved response clarity
Enhanced layout readability
Task Completion
Testing workflows
FINAL SOLUTION
JT Assistant became a context-aware layer within the case management system—helping users access information, complete tasks, and move faster without disrupting their workflow.
Mobile-First Design
Optimized for on-the-go access, ensuring prosecutors can work from anywhere.
Measurable Improvements
Users were able to complete common tasks faster with fewer steps, especially when locating documents or verifying case details.
Testing and early feedback showed a clear reduction in time spent searching and less reliance on manual navigation, with users consistently choosing suggested actions over traditional workflows.