MISTICUS MIND
Turning a fragmented legacy platform into a streamlined legal AI workflow
BACKGROUND
As Founding Product Designer at MisticusMind Inc., I led the redesign of Legal Genius, an AI platform for legal research and drafting. I collaborated across product, engineering, customer success & sales teams, leading end-to-end UX for the product to turn a fragmented legacy experience into a more unified and scalable product.
TEAM
Product Manager
Developer
AI Engineers
Me (Product Designer)
MY ROLE
Sole Designer
UX & UI Design
User Research
Design System
TIMELINE
5 Months
30%
Reduced Task Time
Across core research and drafting flows.
35%
Increase in AI feature adoption
After making AI outputs verifiable and trustworthy.
4 min
Time to first AI query
Down from est. 14 min, 72% faster projected
CONTEXT
Legal professionals spent 30–40 minutes trying to find value - then gave up
Legal Genius is an AI-powered legal technology platform designed to automate legal research, document drafting, and case management for legal professionals.
Legal Genius was built by engineering without UX input and had high UX debt. Users hit the product, got lost, and returned to manual work. The interface made it impossible to discover and reach its full potential. New users were overwhelmed. Power users needed all of it accessible.
This was directly affecting:
-
User trust
-
Productivity
-
AI feature adoption
"I didn't know where to start. It was overwhelming and hard to understand. I couldn't figure out where to click with so many options, so I went back to doing it manually."
- Lawyer, Customer Interview (via CS team)
CONSTRAINTS
Understanding what I was working with:
Before any research or design, I mapped the constraints. These weren't just limitations; they shaped every methodological choice that followed.
CONSTRAINT 01
No direct user access
Confidentiality policies blocked direct interviews. All research ran through internal proxies.
CONSTRAINT 02
High-stakes domain
In legal work, a wrong source or a missed clause has professional consequences. Trust isn't optional.
CONSTRAINT 03
Implementation feasibility for migration
Design decisions had to account for an upcoming developer migration, requiring solutions that balanced improved UX with technical feasibility.
PROCESS
How I got from broken to rebuilt
Full disclosure... I had no idea how the law field worked.
I started with zero legal context. After a lot of Suits and even more 'explain-this-to-me' conversations with my lawyer family, I finally moved past the jargon to understand the actual workflow I was designing for.
Step 01 - Research
I collaborated with stakeholders, product, customer success & sales teams as a research proxy to understand current friction areas, bottlenecks, and requirements; ran a full UX audit; and mapped where lawyers actually are in their workflow before they open Legal Genius.
"Most users who drop off in session 2 say they 'couldn't figure out where things were.' They don't complain about the AI — they never get far enough to use it."
- Customer Success Manager, stakeholder interview
I synthesized this data in Notebook LLM and clustered all the key findings into an affinity map.
Five themes emerged: Information architecture, Cognitive overload, AI transparency, Workflow fragmentation, Onboarding & learnability.
This gave me a shared language with the PM and engineering team, instead of "the sidebar is confusing," we talked about "the Information Architecture theme" with 5 concrete sub-issues behind it.
UX Audit:
The legacy platform suffered from "Technical-First" design, where the system’s architecture was exposed to the user, creating significant barriers to efficiency. My audit focused on three key areas:
-
Navigation & Workflow Efficiency
-
Cognitive Load & Onboarding
-
Human-AI Trust Gap
HOW MIGHT WE
Help legal professionals get to verifiable answers and actionable drafts in the fewest steps without questioning whether the AI can be trusted?
DESIGN RESEARCH
Learning from established & tested design patterns
We met our users where they are rather than teaching new behaviors. By analyzing patterns from commonly used tools like ChatGPT and Claude.ai, which are designed around common mental models, we integrated familiar AI workflows that aligned with lawyers' mental models, reducing the learning curve and boosting tool adoption.
AI first search and nested secondary actionsactions
User control & flexibility within 3 clicks
AI suggestions with sources, document view
Collapsible side nav for easy navigation to key features and recent chats
Showing sources for user trust and verification
Shortcuts for power users
Initial Sketches
INFORMATION ARCHITECTURE
The audit revealed the core problem wasn't just visual; it was structural. The old IA had no hierarchy and had 5+ user actions on the home. The AI lived in a different tab, making it seem like an isolated secondary feature. Research and drafting had no structural connection.
I remapped the entire IA before touching any screen design with my product maanger.
Old information architecture
New information architecture
WHAT CHANGED STRUCTURALLY
The new IA has one primary spine: search → result → case hub → draft → export. Settings become secondary. Context carries forward rather than resetting at each step.
DESIGN PRINCIPLES
Three principles that guided every decision
I reorganised the platform around how lawyers actually think and work, not around features. Each decision came directly from a specific finding.
01
Search is the entry point
The first action should always be asking a question, not navigating a dashboard or filling a form.
02
AI must be verifiable
Every output shows its source. Lawyers need to trace any claim back to the original document.
03
Complex tasks feel guided
Progressive disclosure and clear workflows replace the wall of options that caused paralysis.
Concept exploration & fast validation with Figma Make
I used FigmaMake to generate early concepts to explore different directions and ran them through stakeholder review in a single session. The goal wasn't polish; it was directional alignment. This approach cut validation time by ~50% and gave us clear momentum within days rather than weeks.
❌
Cognitive load high, still too dense, causing the same cognitive fatigue as the original.
❌
Cognitive load moderate. Single search input as the only primary action, but the document grid created a visual clutter.
✅
Matched the new IA direction, following a common mental model inspired by AI tools like ChatGPT and Claude.ai. Easy to navigate side nav and AI search focused with an easy option to create a new case and draft
THE SOLUTION
Decisions that changed how lawyers use the legacy tool
Rather than reorganising features, I reorganised the interface around how lawyers already think and work.
01
PROBLEM
Users arrived at a dashboard with no understanding of what the platform does or how to use it. There was no clear "start here." Most gave up feeling overwhelmed, with everything competing for their attention at once before they'd seen a single piece of value.
SOLUTION
Designed a walkthrough tour for first-time users that explains exactly what the platform does and guides them to take their first action, step by step, without triggering the same overwhelm that caused users to leave. Each step is contextual, brief, and tied to an action rather than a description.
-
Users reached the first value without support
-
Reduced early drop-off at onboarding
-
Time to first time value decreased drastically
02
PROBLEM
Four equally weighted actions competed on the home screen. Users had to think before doing anything. The core AI search, the entire point of the platform, wasn't even visible.
SOLUTION
Made AI search the first entry point. The platform shifts from tool navigation to question-first thinking. Users start a workflow immediately, no orientation required.
-
Immediate task initiation
-
Reduced "where do I start" confusion
-
35% increase in AI feature adoption & user trust
03
PROBLEM
Draft creation required a long form before users saw any value. There was no clear handoff between AI research and drafting, two things lawyers do as one continuous workflow.
SOLUTION
Replaced the form-first flow with a co-pilot drafting interface: the user can directly ask the AI to create a draft in chat, citations auto-populate, and the AI asks a clarifying question before generating, so output is targeted, not generic. Users can switch between AI suggestions and manual editing.
-
30% reduction in task time
-
Research → draft without context switching
-
Users stay in control of output
04
PROBLEM
The interface uses a triple-nested modal loop. This breaks the user's mental model and "Back" button expectations. Every new layer increases the risk of Disorientation
SOLUTION
User can easily access and manage all cases in one place. They can access a 360-degree view of any case, including its overview, documents, drafts, and past AI searches. Assists users in efficiently managing their cases by transitioning from manual management to a unified view where they can find everything and reference past cases in the future.
-
Zero Disorientation Rate
-
+55% Improvement in Feature Discoverability
0-to-1 design system, built in parallel
No system existed when I joined. I built one alongside the redesign, tokens, components, interaction patterns, and engineering documentation. It slowed down early screens and paid back significantly in every sprint after. Full story in a dedicated case study.
80% fewer inconsistencies
DEV HANDOFF
Eliminating the Guesswork for developers
I used detailed annotations to translate high-level UI into build-ready specifications. By explicitly defining input behaviors, button states, and accessibility names directly in the handoff, I ensured that the complex logic of Legal Genius, like AI-assisted inputs, was implemented with zero ambiguity.
IMAPCT
What Changed
30%
Reduced Task Time
Across core research and drafting flows.
Usability Testing
35%
Increase in AI feature adoption
After making AI outputs verifiable and trustworthy.
Analytics Data
4 min
Time to first AI query
Down from est. 14 min, 72% faster projected
Usability Testing
REFLECTIONS
What I'd carry forward
No analytics existed before this project. I instrumented key events and task completions, AI query usage, and draft exports, before launch to establish a clean baseline. Numbers are measured against that baseline.
what Worked
What Worked
FigmaMake for fast concept validation
Three concepts through stakeholder review in one session. ~50% faster than traditional wireframe rounds, and clear direction within days.
Building the system alongside the product
Parallel system-building was harder up front but significantly faster everywhere else afterwards. Getting engineering buy-in from sprint one made it possible.
Do Differently
Still Open
Push harder for even indirect user access
Confidentiality blocked direct interviews. Even 2–3 mediated sessions with real users would have sharpened calls I made on proxy data alone.
AI correction loops
When lawyers edit AI output, that feedback goes nowhere. Building a loop where corrections improve future suggestions is the next meaningful problem to solve.