Amanda
Chan

 


WORK

SOCIAL MEDIA MANAGEMENT

ABOUT

Goldhirsh Foundation - LA 2050

Bringing donors closer to the nonprofits that need them

ROLE
AI/ML Engineer
TIMELINE
August 2024 - December 2024
TEAM
Project Manager
AI/ML Engineers (2)
UI/UX Designer
TOOLS & SKILLS
Python
LangChain
Ollama 
Figma
Natural Language Processing

THE PROBLEMThe LA2050 Ideas Hub had the right resources, but they were difficult to find.
 LA2050 is a community-guided initiative launched by the Goldhirsh Foundation in 2011 to drive civic progress across Los Angeles. Over the past decade, it has engaged hundreds of thousands of Angelenos, awarded grants to 254 winning proposals, and built an Ideas Hub that connects residents, organizations, and policymakers around five shared goals: to make LA the best place to connect, create, learn, live, and play.

That ambition had a navigation problem. The Ideas Hub hosts hundreds of nonprofits, for-profit initiatives, and community proposals — but exploring it required users to already know what they were looking for. Donors and volunteers without a specific organization in mind had no clear path to discovery. 


RESEARCH INSIGHTS Users don't search for nonprofits the way they search for products. Intent is vague, emotionally driven, and conversational. A keyword-only search wasn’t realistic. Instead of just matching strings, it needed to understand meaning. The solution required meeting users where they were, in natural language.

PROCESS I came into this project as a developer and left thinking like a product person.

My contributions spanned data preprocessing, LangChain integration, and the visual design of the chatbot interface. Working across both the technical pipeline and the user-facing layer gave me a clear view of how backend decisions shape front-end experience. 

  • Python — cleaned, structured, and explored the Ideas Hub dataset to prepare it for search and retrieval

  • LangChain + Ollama — integrated language model tooling to enable natural language understanding and response generation

  • Figma — designed the chatbot interface, translating technical output into an experience that felt approachable and intuitive for donors and volunteers


DESIGN DECISION
Combining two search methods to handle the full range of how people ask questions.
 A keyword-only search fails the user who wants to explore, while a semantic-only search can miss specific details. To solve this, we opted for a hybrid approach — allowing the chatbot to handle both precise queries and open-ended questions within the same interface. This helped close the gap between what a user means and what the system returns.


DESIGN DECISION
Surfacing trusted organizations first to reduce decision fatigue.
 With hundreds of results possible, returning an unranked list would have recreated the same friction the chatbot was built to solve. Prioritizing Goldhirsh Foundation award recipients, organizations with multiple proposal submissions, and IRS-compliant nonprofits gave users a confident starting point rather than another wall to sort through. This was the moment I most clearly saw how product decisions shape user trust.


DESIGN DECISION
Designing a chat interface that felt familiar, not technical.The LA2050 Ideas Hub serves a broad audience: donors, volunteers, community members, and nonprofits themselves, ranging widely in how they interact with technology. Designing for the most digitally fluent user would have left everyone else behind.

As a result, I built the user experience around a conversational UI pattern with familiar iMessage-style bubbles and suggested prompt chips for users who weren’t sure where to start. A persona, Sunny (whose name comes from LA’s famous sunshine), made the experience feel like an interaction rather than a query. Additionally, the clarifying question flow with a focused follow-up kept the experience guided and manageable at every step. 
 


SOLUTION The LA2050 Navigator made a dense civic database feel navigable and human.

By combining natural language understanding with intelligent prioritization, the chatbot connected users to the right organizations without requiring them to know exactly what they were looking for. The result was a tool built to amplify giving within LA.



REFLECTION
This project clarified the kind of work that I’m passionate about. Coming in with a software development background, I expected the technical problem to be the most engaging part. What held my attention more was the layer above it — understanding why we were building what we were building, who it was for, and how each decision shaped the experience on the other end. Navigating that across a team of five pushed me to think beyond implementation and toward product thinking.

What I’d do differently I'd get more intentional about the process earlier. Managing contributions across a five-person team while keeping the user goal in focus was harder than the technical work itself. That friction — and the clarity that came from working on both the pipeline and the interface — pointed me toward product design and management. The problems I want to solve live at the intersection of people, systems, and experience, not just code.

© 2026
AMANDA CHAN