: Improving Case Law Search Accuracy Through UX-Driven User Articulation
A complicated case law search system
for individuals with limited legal knowledge
Confronting the complexities of a personal legal case through natural language processing (NLP)
When AI Why does 'Specificity of Details' matter?
During the closed beta test, results showed that more detailed user statements were correlated with higher relevance and applicability of the retrieved case law.
The key details vary by case type. Given the differences across the 10 categories, I divided them into 10 subcategories systematically gathering essential factual elements for a total of 100 distinct legal issues.
Gathering essential key details for a total of 100 distinct legal issues for Query Database
1/2
Legal situations are often more sensitive than they appear, where small details can affect the outcome.
Lawlow, AI-powered Legal Case Search Engine
2023-24 Work Experience
My role
brief context
Team
Product Designer
Where It All Began
technical Approach
Key UX Factor
How I designed the query for an AI-guided recommendation system
Concept Development (30%)
Branding (80%)
Usability Test (70%)
UX Data Analysis (70%)
Design System (100%)
Interface Design (100%)
Interaction Design (100%)
Prototyping (100%)
Responsive Design (100%)
Product Designer (Me)
Established
Start-up
3 Engineers
Marketer
Solely responsible for designing all 4 main pages
and 21 pop-ups, ensuring full responsiveness,
complete user flows, and a cohesive design system
For these reasons, users struggle to find the best way to search for similar cases based on their own situations.
To address this, we introduced an LLM-based system that allows users to describe their situations in natural language, enabling critical details to emerge more naturally than with form-based inputs.
I evaluated satisfaction based on 472 responses from 268 users during the closed beta. Statement specificity was measured based on whether three types of information were present in each statement: cause-and-effect relationships, time and place, and additional contextual details.
For example in assault cases, the verdict can change depending on where the person was struck, the force used, and the context of the incident. This is why we have focused on these critical details, as they play a decisive role in aligning with the most relevant case findings.
- AI Engineer
- Front-end Engineer
- Back-end Engineer
Duration
Tools
Dec 2023 - Present
Figma
Responses with less specific
questions
60
%
This represents a 42% relative increase in satisfaction, indicating a substantial improvement when users provided more specific statements.
42
%
Satisfaction (%) on relevance exploration & applicability of detailed legal provisions
Responses with more specific
questions
85
%
Lists of Users (Admin page)
Lists of Statements (Admin page)
Sample-Based Learning &
AI-guided Comments
Outcome-to-Insight
Interpretation System
UX Solution
Focus Areas in UX
Improving Input Quality
Gaining an Objective Sense of Abstract Qualities Through the Sample
Focused Search Recommendations for Gaining Specialized Insights
Controlling actual writing through
AI-guided recommendations
Efficient Case Comparison Based on Similarities and Differences
The sample serves as an objective reference point for abstract qualities like 'detailed explanations' and 'a casual tone,' helping users learn effectively and help them to take practical action.
Within a comprehensive collection of highly relevant case precedents, users can utilize specialized keyword search options to gain deeper insights into the specific aspects they seek to explore.
It provides real-time feedback on missing key details, offering writing direction. While sample-based learning lacked control, AI ensures practical specificity by identifying gaps and guiding the writing process.
Access key information without reading lengthy case documents by comparing my situation with similar cases. The system highlights relevant details worth exploring, naturally guiding focused searches.
solution 1
solution 1
solution 2
solution 2
Case law is not just a reference but an essential tool for navigating legal challenges. Although access has improved, finding relevant cases remains a challenge for most people due to the vast amount of data and a lack of understanding of legal terms and case types.
Lawlow is a Seoul-based legal-tech startup focused on case law search. In this project, I improved the user experience by guiding users to articulate their legal situations in greater detail, resulting in more relevant case law outcomes.
General Legal Expression
Battery resulting from intervention in a domestic dispute
I was just trying to stop a fight between a couple and the guy suddenly punched me in the stomach
Expression with Natural Language
1. Keyword Uncertainty
Users only encountered limited and fragmented results depending on the specific keywords they searched for.
Users didn’t know where to stop reading and spent a significant amount of time endlessly scrolling through results.
Users were unsure which category their case belonged to, forcing them to look for clarification elsewhere.
Why problem?
Why problem?
Why problem?
2. Result Overload
3. Categorization Confusion
jaywalking-related vehicle accidents
= Jaywalking accident liability
= Pedestrian fault car accident
= Crosswalk violation vehicle collision case
= Jaywalking traffic accident court case
The Civil Rights Act of 1964
The Fair Labor Standards Act (FLSA)
The Americans with Disabilities Act (ADA)
The Clean Water Act (CWA)
The Foreign Corrupt Practices Act (FCPA)
The Voting Rights Act of 1965
Displaying page 1 of
Total Results for "Crosswalk violation"
59,409
594,085
How might we encourage users to provide more detailed and accurate factual statements?
Improving Input Quality
Improving Outcome Interpretability
How might we help users synthesize insights from case law to better anticipate the potential outcomes of their own cases?
Launched
Design Iteration
4 Factors Evaluated for Effectiveness
Through Usability Testing
There were two main considerations for each of the 1 and 2 solutions during the decision-making process. Using Figma prototyping tests, I compared the effectiveness of each option with feedback from four users.
Dual Mode with Guided Autonomy
Within Solution 1
Within Solution 2
Efficient Information Layout
Sample Display Methods
Case Law Structure and Display Order
A
C
B
D
A
Dual Mode with Guided Autonomy
B
Sample Display Methods
C
Basic Layout Structure for the Results Module
D
Efficient scanning of key information
Free Writing Mode
A - no sample
A - CTA on top
B - random sample
B - fragmented placement of elements
C - random sample by topic
C - A clear flow of title-content-CTA
Guided Writing Mode
- An issue arose where the options meant for viewing samples were mistakenly perceived as mandatory fields for describing one's situation
- Testing revealed no significant differences in statements between the two modes. Instead of prior legal knowledge or past consultation experience,
the key variable was the understanding and learning of the narrative direction. Based on this, I decided to unify the modes.
Word count varies significantly among individuals, with some writing as little as one sentence
The CTA button showed the lowest engagement because the visual flow led users’ attention downward after encountering the button, causing it to be overlooked.
The fragmented placement of elements such
as the ‘See Original’ button, relevance score creates a disjointed visual hierarchy, disrupting the user’s natural flow of attention.
While the word count range has narrowed, many still write concisely, and there's a notable difference in specificity
The information lacks a clear hierarchy, presenting all elements at the same level, which makes it challenging for users to quickly scan and grasp the key points.
Responses are generally over 80 characters, with higher specificity, as users tend to provide more details when selecting topic their situation
Placing the title alongside the relevance score ensures clear communication, while the title-content-CTA flow creates the most stable visual hierarchy.
I initially created two modes with varying levels of guide intervention based on users' familiarity with the situation and legal knowledge. Experienced users can freely describe their situation, while new users receive step-by-step guidance with key details to help them complete each section.
I gathered four testers and asked them to provide responses in sequence (A, B, and C). I observed a noticeable difference in both the quantity and specificity of their answers. As I progressed from A to C, the responses became progressively more detailed and insightful.
I focused on designing a foundational layout structure for the results module that
guides the most stable visual flow and effectively boosts CTA button click-through rates.
Since users will be viewing multiple similar case precedents, it was important to enable them to quickly grasp the key points before reading the full details of each case. Therefore, I focused on the visual hierarchy of the information.
The defendant is convicted of misappropriating funds meant for Company E, in violation of the Improper Solicitation and Graft Act. Their actions, including diverting funds and fleeing abroad, reflect an intent to escape legal consequences.
The defendant operated Company E in Gangnam and managed Company F from 2012 to 2016. In 2014, the defendant introduced the "G" membership plan. Between December 2016 and January 2017, the defendant misappropriated over KRW 2.2 billion, fled to Thailand, and ceased operations.
The Court finds the defendant guilty of misappropriating funds intended for Company E's operations, thereby violating the Improper Solicitation and Graft Act. The defendant’s actions, including the misuse of funds and subsequent overseas flight, demonstrate intent to evade legal accountability.
Problem
Facts
Judgement
#1 Golf Membership Sales Company Misappropriation of Funds and Overseas Flight Incident
Relevance
75%
Access the Original
Compare with My Case
The defendant is convicted of misappropriating funds meant for Company E, in violation of the Improper Solicitation and Graft Act. Their actions, including diverting funds and fleeing abroad, reflect an intent to escape legal consequences.
The defendant operated Company E in Gangnam and managed Company F from 2012 to 2016. In 2014, the defendant introduced the "G" membership plan. Between December 2016 and January 2017, the defendant misappropriated over KRW 2.2 billion, fled to Thailand, and ceased operations.
The Court finds the defendant guilty of misappropriating funds intended for Company E's operations, thereby violating the Improper Solicitation and Graft Act. The defendant’s actions, including the misuse of funds and subsequent overseas flight, demonstrate intent to evade legal accountability.
Problem
Facts
Judgement
#1 Golf Membership Sales Company Misappropriation of Funds and Overseas Flight Incident
Relevance
Compare with My Case
See Original
The defendant is convicted of misappropriating funds meant for Company E, in violation of the Improper Solicitation and Graft Act. Their actions, including diverting funds and fleeing abroad, reflect an intent to escape legal consequences.
The defendant operated Company E in Gangnam and managed Company F from 2012 to 2016. In 2014, the defendant introduced the "G" membership plan. Between December 2016 and January 2017, the defendant misappropriated over KRW 2.2 billion, fled to Thailand, and ceased operations.
The Court finds the defendant guilty of misappropriating funds intended for Company E's operations, thereby violating the Improper Solicitation and Graft Act. The defendant’s actions, including the misuse of funds and subsequent overseas flight, demonstrate intent to evade legal accountability.
Problem
Facts
Judgement
Golf Membership Sales Company Misappropriation of Funds and Overseas Flight Incident
#1
Compare with My Case
A - No hierarchy
B - Clear hierarchy
By reducing the information to two key categories and differentiating visual levels based on importance, users can more quickly grasp the core points.
Takeaway
This project showed me that even small, modular UX changes can meaningfully impact user trust and ultimately translate into measurable business outcomes.
Through A/B testing, I learned that how users are guided to articulate their situations, how detailed the prompts are and how examples are structured directly shapes both the quantity and specificity of their input.
More importantly, higher-quality input doesn’t just improve data quality. I realized it fundamentally shapes users’ expectations of the output, directly influencing case law results and overall service satisfaction.
Simple assault
(Criminal Law Article 260)
Justified self-defense
(Criminal Law Article 21)
Assault causing bodily injury
(Criminal Law Article 257)
Aggravated assault
(Criminal Law Article 261)
Assault Severity & Area
Assault Context
A
A
B
B
A drunk person bumped into another person’s shoulder on the street, leading to a brief altercation and a small bruise on the arm.
A man walking down the street was approached by someone who made threatening remarks, and the man struck the individual
A person was punched in the face during a street altercation,
causing a broken nose and requiring medical treatment.
A person initiated an unprovoked attack on another individual at an intersection, injuring them in the process.
Improving Outcome Interpretability
[ Example of Assault and Battery]
Parties Involved
Witnesses or Evidence
Date, Time, and
Location of
the Incident
Incident Details
Physical Actions
and Injuries
-
-
-
Injuries or Damages
Evidence
1
2
3
5
4
Common & Default Items
Detailed & Alternative Items
Principles of Item Generation by Type
in AI-Guided Recommendation Systems
A set of five description recommendation items will be constructed by first identifying five basic common items. Among these, two items will be replaced with specific key factors that align with subtypes, ensuring a total of five tailored items in the final set.