.png)
Infilla
AI-enhanced search to support planners' decision-making
CLIENT
INDUSTRY
Infilla
Civic Tech, Urban Planning
My Role
Product Designer
Team
6
Responsibilties
Research, UX & Prototype Lead
Timeline
10 weeks
Tools
Figma, Figma Make,
Lovable
What is Infilla?
Infilla is a civic-tech company that helps U.S. city planning departments streamline their operations through multiple digital tools. Infilla's flagship product, the Forum, is a centralized Q&A and knowledge-sharing platform designed for planners, zoning staff, and citizens. It enables users to ask regulatory and zoning questions, access expert answers, review past discussions, and find verified resources in one place.
Users
Our primary users are permit counter staff who interact directly with applicants under significant time pressure. They need quick access to reliable, verified answers on zoning codes, land use regulations, and development procedures that they can confidently share in real-time.
Problem
Planners couldn't find the regulatory answers, zoning codes, and permit procedures in the forum they needed to answer to applicants due to:
-
Poor search accuracy: Search couldn't handle keyword variations like capitalization ("ADU" vs "adu") or formatting ("SB9" vs "SB-9"), returning either zero results or too many irrelevant results.
-
Manual result navigation: Planners had to open each forum answer to find sources (Laws/permits) and verify for themeselves reading through whole documents to determine if that was what they were looking for.
-
Complex documents Interpretation: Junior planners had difficulty interpreting complex documents and needed the timely help of senior experts.
​Search data confirmed that the forum was underperforming, with 38.5% search failures, 12% queries returning zero results, and 61% of searches creating information overload.
User Problem
-
Lost time during time-sensitive interactions
-
Frustration from repeated searches and duplicated question posts
-
Reduced confidence in the system
Business Problem
-
Potential decrease in platform usage
-
Underutilized knowledge in the system​
DESIGN CHALLENGE
HMW design an AI-supported search experience that clearly reveals where answers come from, builds trust through transparent evidence, unifies information from multiple sources & supports users in making confident, well-informed decisions?
Solution
We designed an AI-enhanced search that understands keyword variations, surfaces relevant sources under horizontal navigation, and guides users through answer excerpts in forum discussions and highlights for external documents in the side panel, helping them improve search precision and navigate seamlessly across sources, ultimately supporting faster decision-making.
Prototype demo- pre-interview onboarding flow covering device check, browser permissions, and error states
RESEARCH
Exploring AI search solutions across platforms
Before designing for Infilla, we wanted to understand how AI search is evolving across products. Since there were very few examples in civic tech, we looked at adjacent platforms across industries to see how LLM-powered search is being used and what we could learn from them.
This helped us define clear principles for designing AI-enhanced search in a civic environment and observe how do they:
-
Add context through filters and prompts
-
Surface answers with supporting sources
-
Present AI summaries responsibly
-
Handle edge cases where clear answers don’t exist​
Transparency is Non-negotiable
Platforms like Gemini and Perplexity surface sources and citations clearly, helping users understand where information comes from and verify it easily.
Guidance Improves Usability
Tools such as Reddit Answers and NYC’s MyCity Chatbot help users refine queries, explore related topics, and navigate toward relevant answers faster
Trust and Safety are Critical
Platforms like GovAI and Alberta AI clearly distinguish expert-verified responses from AI-generated interpretations and limit sources to credible, approved information to reduce hallucinations and misinformation.
Designing Early Concepts on Assumption
Taking inspiration from competitor patterns and the principles defined for responsible AI, we initially explored a chat-based interface.
.png)
Planner enters a zoning question in the AI chat interface.
Our Assumptions
Since Infilla already had a knowledge base, we assumed a conversational AI experience would allow planners to ask open-ended questions and receive instant answers with clear natural-language interpretations with in-line citations, helping them find information quickly.
AI returns a synthesized answer pulling from sources with links for quick verification.
Survey to Understand Planners' Search Process
From client conversations, we knew planners relied on keywords, but as we began designing filters and conversational search, several questions emerged: How do planners actually search using just keywords? When do they use filters? What are they looking for, and where is the search currently failing?
After several back-and-forth searches in existing forums ourselves. We needed answers to these questions so we could ground our design decisions in real-world behavior. We ran a survey to better understand planners’ search processes, forum usage, key pain points, and perceptions of AI. This helped us see search through the planners’ lens rather than rely on assumptions.
What Survey Revealed
100%
of planners expressed distrust in AI-generated interpretations. Why?
Planning decisions carry legal, financial, and professional consequences. Planners cannot risk AI misinterpreting nuanced zoning regulations.
How Planners Search, Trust and Decide?
Our survey revealed that: planners avoid AI interpretation, verify independently, and prefer staying in control of judgment.
AI cannot interpret nuance reliably
Complex planning and zoning codes require human judgment and contextual understanding. AI can hallucinate.
Independent verification is critical
Planners cross-check official codes, consult colleagues, and review past cases even when answers are expert-reviewed.
Self‑sufficiency matters
Planners want control over interpretation. They see AI as support for research, not a replacement for judgment.
The AHA-Moment
Planners need a research tool — not an AI assistant; Planners don’t trust AI to interpret regulations. They want AI to surface reliable results faster so they can review and verify information themselves.
“I would not trust AI to interpret the Planning Code. I would trust it to explain how a word or phrase is used in the overall context so that I can make my own determination.”"
Senior planner, Sanfrancisco
THE PIVOT
From AI assistant to Invisible AI search
The survey insights led us to change direction. Instead of building an AI that generates answers or interprets laws, we moved to an “invisible AI” approach that improves search while keeping planners in control.
AI works quietly in the background to understand keyword variations, surface relevant sources, and highlight useful information so planners can find what they need faster without AI making decisions for them.
Our guiding principle was simple: interpretation and decision-making around complex laws should remain human, while AI should support the research process.
To support this shift, we focused on designing an AI-enhanced search experience that prioritized transparency, control, and verification.
Exploring Early Directions After The Pivot
We translated our strategic shift into early concept explorations for an AI-enhanced search experience supporting keyword, filter, and natural-language queries. Key design decisions we explored where AI quietly works in the background to:
1
Surface all relevant sources in a horizontal tab
A horizontal navigation surfaced municipal code sections, external documents, and forum discussions in one place, reducing the need to dig through multiple threads manually.
2
Show updated, relevant results under each source
Each source displayed the most relevant and recent results, allowing planners to quickly scan and decide which to open.
3
Pull expert-backed answers from forum discussions
Instead of generating new AI answers, the system surfaced excerpts from trusted expert responses already available in the forum.
3
Highlight and cite external documents
For external codes and documents, AI provided highlights and direct citations rather than interpretations of nuanced laws, enabling quick scanning while preserving accuracy and trust.​
We ensured consistency with the current design system. Even though we ditched the conversational AI approach, AI patterns discovered in that phase were carried forward—inline citations, expandable source panels, and suggested follow-up questions.
DESIGN CONCEPTS
AI Enhanced Search
for search precision
This concept focuses on giving planners more control through semantic search that recognizes keyword variations like hyphens, abbreviations, and capitalization, alongside contextual filtering and smarter search refinement.
Feature #1
Suggested Filters
We introduced suggested filters so that when a planner types a search, the system automatically suggests relevant filter options, such as zoning categories. This reduces cognitive load for planners compared to manually selecting each filter individually. As a result, planners spend less time scrolling through irrelevant results.
Camera preview. Device Selectors.
Feature #2
Additional Context Filters
Additional filters such as "Add Attachment" and "Add Address" were introduced, as planners often rely on project-specific details when researching zoning questions.
Allowing users to include site addresses or supporting documents directly within search adds important context and improves the relevance of surfaced results.
Camera preview. Device Selectors.
Feature #1
Unifield layout for
Result Navigation
Previously, primary sources like municipal codes and official documents were scattered within forum threads or buried behind filter menus, forcing planners to manually open and scan multiple posts to find relevant information.
We introduced a horizontal source navigation that consolidates forum discussions, municipal codes, and external documents into one unified view. When planners search, each tab surfaces the most relevant and updated results within that source. This means quick scanning, easy comparison, and reliable information without digging through scattered threads.
Camera preview. Device Selectors.
DESIGN CONCEPTS
Unified layout for result navigation and side panel for source visibility and verification
This concept focuses on giving planners more control through semantic search that recognizes keyword variations like hyphens, abbreviations, and capitalization, alongside contextual filtering and smarter search refinement.
Design Concepts
Unified layout
for results navigation n
Previously, primary sources like municipal codes and official documents were scattered within forum threads or buried behind filter menus, forcing planners to manually open and scan multiple posts to find relevant information.
We introduced a horizontal source navigation that consolidates forum discussions, municipal codes, and external documents into one unified view. When planners search, each tab surfaces the most relevant and updated results within that source. This means quick scanning, easy comparison, and reliable information without digging through scattered threads.
Design Concept
Developing on to our sketches, we focused on two core concepts to support planners’ research workflows and decision-making.
Concept 1: AI Enhanced Search Experience for search precision
This concept focuses on giving planners more control through semantic search that recognizes keyword variations like hyphens, abbreviations, and capitalization, alongside contextual filtering and smarter search refinement.
Suggested Filters
We also introduced suggested filters so that when a planner types a search, the system automatically suggests relevant filter options, such as zoning categories. This reduces cognitive load for planners compared to manually selecting each filter individually. As a result, planners spend less time scrolling through irrelevant results.
Additional Context Filters
Additional filters such as "Add Attachment" and "Add Address" were introduced, as planners often rely on project-specific details when researching zoning questions.
Allowing users to include site addresses or supporting documents directly within search adds important context and improves the relevance of surfaced results.
Concept 2: Unified: Unified layout for results and side panel for source visibility and verification
This concept focuses on source visibility and confidence in verification.
Unifield layout for Result Navigation
Previously, primary sources like municipal codes and official documents were scattered within forum threads or buried behind filter menus, forcing planners to manually open and scan multiple posts to find relevant information.
We introduced a horizontal source navigation that consolidates forum discussions, municipal codes, and external documents into one unified view. When planners search, each tab surfaces the most relevant and updated results within that source. This means quick scanning, easy comparison, and reliable information without digging through scattered threads.
.png)
Side Panel Preview
Previously, planners wasted clicks opening each source and wasted time reading entire documents to check if they were relevant. We designed side panel previews for forum discussions and documents. When a planner clicks a result, a preview opens alongside search results showing excerpts of expert answers for forum discussions and AI generated highlights with key citations for documents. Planners can now quickly assess relevance, verify sources, and decide what to open without leaving the page.
.png)
Concept Testing & Iteration
We conducted concept testing with five planners across departments to evaluate the enhanced search experience. We aimed to understand whether planners would trust AI-surfaced results, navigate across multiple sources with ease, and find the information accurate and useful. We also assessed if the design reduced the effort required to search and interpret information while supporting confident, independent decision-making without AI feeling intrusive.
Key Findings
-
Search behavior varied across planners. Some searched with keywords, others with filters, and many switched between both. This validated our flexible search design that lets planners query however they prefer.
-
Planners strongly valued clear source separation (forum, local code, state law, documents), answer previews for quick scanning, and the side panel that lets them view answers and code references without losing their place in search results.
-
100% said the side panel would significantly improve their workflow. They provided feedback on content hierarchy, layout, and usability to reduce cognitive load and search friction, which informed our final design iteration.
-
Less tech-savvy users needed onboarding. Planners suggested guidance on new features to help users who aren't as comfortable with technology.
Iterations
-
Improved visual hierarchy in the side panel with clear sectioning for questions/answers/tags
-
Highlighted key answers in the forum answers excerpt
-
Added a prominent disclaimer that AI highlights support research, but does not interpret nuances of laws
-
Onboarding flow of new features for less tech-savvy planners
Impact & Result
Testing validated the design across all stakeholders.
-
​All five planners confirmed the enhanced search would significantly improve their research workflow.
-
Engineering approved technical feasibility.
-
Infilla is now rolling out these features to production.
Next Steps
-
Explore tailoring the experience for an external, public-facing forum for citizens
-
Explore opportunities to detect outdated data and assess the technical feasibility of using AI to identify outdated information.