GovScan
Designed an LLM-based web application to help policy analysts find specific data within reports, saving them about 80% of the time*.
​
* Time spent in manually sifting through documents
My Role: UX Research, UX Design
Duration: 6 weeks (Research - MVP)
Sponsor: United States Digital Service
Team:
Tyler F - Project Manager
Davis C - Developer
Eashwari S - UX Researcher
Context
Program reports are compliance documents (~300 pages) that show the federal government how the states run federal funded programs. These PDF documents explain the details of how the state manages these programs within their state. Yet these resources are incredibly difficult to use as a policy analyst because you can't easily compare across states or parse the information within.
Outcome
Reduced 80% of the time taken by policy analysts in sifting through large PDFs.
Solution
A LLM based web platform to search specific information in program reports and compare it across different states.
Problem
Policy analysts struggle to effectively use program reports because of its vastness and the time constraints.
Research
But how did we reach here?
The entire research was divided into two phases
Discover
The first phase was all about user research and understanding their workflows.
Methods Used:
User Interviews, Affinity Mapping, JTBD, Personas, Journey Mapping, Storyboarding.
Define
The second phase was about translating findings into product opportunities.
Methods Used:
2X2 matrix, Design principles, POGs, Diagram.
Understanding the problem
The 12 interviews with the policy experts echoed a common sentiment. They were all looking for specific information.
What was even more surprising is that the main job of all types of policy analysts was not to analyze reports but to make informed decisions or recommendations.
​
"Finding a specific information in these reports is like finding a needle in a haystack".
​
"How do I evaluate whether the report is worth reading or not?"
​
Who are the primary users?
Through our research we analyzed the 3 types of policy analysts Federal, NGO and State level. However, analysts at federal level have to read and analyze reports from 50 different states for 2000+ federally funded programs. ​
Aha! moments
After a thorough analysis of interview findings we understood why policy analysts find it difficult to navigate through program reports.
There is a mismatch in the existing mental models of two user groups - one who generate reports and ones who consume them.
​
Different states satisfy the same federal regulations in different ways.
​
Design opportunities
-
How might we help policy analysts to quickly find specific information within program reports?
-
How might we help policy analysts to easily compare variations in specific parts of reports across states?
Defining the solution space
The conclusion was that any solution we develop has to be positioned such that it does the laborious work of finding and parsing data, while leaving the high-risk analysis in the hands of humans. For this particular problem having a human cross-check everything that the tool puts out is extremely crucial for our target users.
Ideation
Building the solution
The phase was more towards getting a high-level idea of the solution.
Exploring the breadth
After 3 hours of brainstorming and discussion session we were finally able to get a macro-level idea of the various possible solutions.
Exploring the depth
We then used future-state storyboarding technique to go deeper into how our solution would fit in user's existing workflow.
Early prototype & testing
Based on the above storyboard we created an initial functional interface prototype. We decided to test it early and get user's feedback. This feedback later became the guiding light for the final interface design.
Learnings from user testing-1
Based on the user feedback we tweaked the existing prototype. For this iteration we added 2 program reports to understand how the users will interact in comparison mode.
Learnings from user testing-2
Brainstorming layouts
Based on the valuable insights from user tests, I brainstormed various interface layouts within technological and logistical constraints.
Solution
The final product
GovScan is an AI web platform which helps policy analysts summarize, compare and find specific data from extensive PDF documents within seconds.
Learnings and Takeaways
Learnings
I learnt that conducting user testing at an early stage is incredibly advantageous and cost-effective. By involving users early on, we can make informed design decisions, resulting in a more efficient final product.
I've come to understand that having a team with people who know different things is really important when dealing with messy problems.
If I have to revisit ...
I would have increased the number of user tests, examining how the product would perform across various types of reports. Additionally, I would work on crafting a comprehensive product launch strategy, prioritizing features and defining success metrics for the product.
View other projects
