MyCity

Project overview

MyCity is a full-stack project I built solo at the UC Berkeley AI Hackathon 2025, held on June 21–22. The prototype demonstrates a fully functional end-to-end workflow, covering data crawling, content generation, database integration, and a frontend web app. See link to my final submission here.

Inspiration

Local news deserts have been expanding rapidly across the U.S., with 3,300 newspapers closing over the past two decades. While short videos on emerging platforms offer quick updates, the MyCity app series serves as a reliable and accurate source of truth, helping people stay informed about what’s happening in their own cities.

What it does

MyCity is a web app—soon to be primarily mobile—that crawls selected official information sources at carefully chosen intervals to deliver timely updates. Drawing on the expertise of seasoned journalists to identify authoritative sources, the app uses one-shot and few-shot prompt engineering to guide its professional writing. The result is relevant, bite-sized, and unbiased information briefs—free of fluff.

Stack

Python, Claude API, SQLite, FastAPI, Node.js, React, and Vercel.

Key challenges

The two secret ingredients of this project—gathering meaningful information based on lived human experience and producing strong content—also posed the biggest challenges. Properly guiding LLMs in professional news writing requires extra onboarding time. For instance, when condensing lengthy documents such as city council meeting minutes, considerable effort went into prompting the Claude API to generate high-quality summaries. Well-crafted few-shot prompts were essential in these cases.

Accomplishments

The current prototype includes content relevant to commuters, such as traffic updates and major public transportation alerts. In the Fremont city prototype, the current workflow can handle the daily output of more than one full-time journalist already. In a newsroom setting, this tool would allow professional journalists to focus on more creative and investigative tasks.

What's next for MyCity

An ideal plan would include expanding the Fremont MyCity prototype with additional local sources and features. Another section of the app will allow users to receive updates from various local event calendars. Users can like or skip events to build a personalized MyCity event calendar. In the next phase, we will also develop a mobile-first version of the app. After that, it would be feasible to scale the app series and bring local news services to small cities across America.

Steps finished by the end of the hackathon

• Web crawling workflow set up
• Prompt engineering done for the writing of select topics
• Database created
• Working API connected to the database
• Basic frontend built

• Set up tutored writing via Claude API

Example text:

The Fremont City Council approved several major infrastructure and service spending items totaling over $9 million at their May 20 regular meeting. All items on the agenda passed unanimously during the meeting.

Key Items Approved:

• Road Repairs: $4.986 million contract for the 2025 Cape and Slurry Seal Project to maintain city streets
• Parkland Purchase: $800,000 to buy property at Fremont Boulevard & St. Leonards Way for parks
• Street Maintenance: $1.03 million annually for 7-year landscape maintenance contract for city streetscapes
• Equipment Purchases: $836,000 for two new street sweepers; $245,000 for Wi-Fi upgrades at city facilities
• Fuel Costs: $1.463 million annual contract for gasoline and diesel fuel for city vehicles
• Transportation Services: $510,000 annually for Lyft ride-hailing services and $120,000 for call center services for seniors and disabled residents through 2028

Mock main page: