Overview
The Ultimate Boston Guide is an interactive guide to Boston and the surrounding areas, created for my final project in my AI and Economics course at Northeastern University . It brings together restaurants, cafes, museums, parks, photo spots, historic sites, shopping, entertainment, hidden gems, and day trips into one place, all displayed on an interactive map with filters for category, neighborhood, and price. The goal is to build something that feels like getting recommendations from a local friend: specific, practical, and easy to explore.
I am building the entire project in collaboration with Claude Cowork, an AI agent that helped me research places, write code, and style the site. It is my first time building a full product with an AI as a true partner, and it has completely changed how I think about what one person can create.
**Keep in Mind: This case study and project is a Work in Progress!**
The Goal: Scale and Accuracy Together
From the start, I wanted the guide to be large and accurate. Most city guides either cover a handful of famous spots or list hundreds of places with outdated or made-up details. I set out to build something that could grow to hundreds, and eventually thousands, of places across Boston neighborhoods, nearby cities like Cambridge and Somerville, and day trips like Salem, Concord, and Newport.
To get there, I set clear goals:
Comprehensive Coverage: Include a wide range of categories and areas so that every kind of visitor, from students on a budget to families to couples planning a date, could find something useful.
Trustworthy Information: Every place had to be real, open, and correctly located, with descriptions written in original language rather than copied from reviews.
Easy Exploration: Make browsing feel effortless through a map, filters, and themed tags like budget, rainy-day, date-idea, hidden-gem, and student-favorite.
Personalization: Let users save their favorite places, add their own recommendations, and plan out their day.
Working With AI as a Teammate
The biggest shift in this project was learning to treat AI as a collaborator rather than a tool. Instead of writing every line of code myself, I wrote a detailed project guide (a CLAUDE.md file) that acted like a creative brief and rulebook for the agent. It defined the data format for every place, the categories and neighborhoods, the writing style, and a set of hard rules, like never inventing facts and never copying text from other sources.
Claude Cowork then took on the job of finding gaps in the guide (for example, a neighborhood with very few cafes), researching candidates, verifying each place against its official website and another independent source, and adding it to the database in batches. A validation script checked every batch for problems before anything was saved. I even have multiple agents work in parallel on different neighborhoods and merge their results.
This setup taught me that the quality of AI output depends heavily on the quality of the direction you give it. The clearer my instructions were, the better and more consistent the results became.
Designing the Experience
Styling was one of the most surprising parts of this project. I’ve come to love design, and working with AI made it incredibly easy to bring my visual ideas to life. Instead of spending hours tweaking CSS, I could share images and examples of what I wanted the site to look like, and Claude would translate them into real, working styles. If something didn’t feel right, I could describe the change in plain language and see it updated almost instantly.
This made the design process feel much more like iterating in Figma than writing code. I could experiment with layouts, colors, and card designs quickly, and spend more of my energy on decisions about the user experience rather than implementation details. The final site is clean and easy to read, with a list of place cards next to an interactive map, and it adapts to a mobile layout on smaller screens.
More Features! Adding Accounts and Saved Places
Once the core guide was working, I realized how much more I could build. One of the biggest additions was user accounts. Anyone can browse the guide without signing up, but creating an account lets users save places with a heart button and add their own recommendations.
Each user has a profile page showing their username, when they joined, their saved places, and the places they’ve added. Clicking any of them zooms the map straight to that location. I also learned a lot about security along the way. Passwords are never stored in plain text, sessions are checked on the server for every protected action, and users can only change their own saved lists. Understanding why these rules matter was one of the most valuable lessons of the project.
Next Steps: Plan a Day
The feature I’m currently working on is Plan a Day. Rather than just browsing a long list, users will be able to put together a full day in Boston, combining places like a morning cafe, a museum, a lunch spot, and an evening activity into one plan. It turns the guide from a directory into something people can actually use to organize their time, whether they’re a tourist visiting for the weekend or a student looking for something to do on a Saturday.
Right now I’m working through how the experience should feel, from how users choose and order their stops to how the finished plan shows up on the map so they can see their whole day at a glance. Building this feature is pushing me to think beyond individual places and design for how people actually move through a city.
What I learned
Building with AI is powerful, but it isn’t effortless.
Keeping Information Accurate: AI can confidently produce details that sound right but aren’t. I’ve built accuracy into the process itself, with strict rules against inventing facts, a verification step requiring two sources, and a validation script that catches errors before they reach the site.
Keeping It Simple: With AI, it’s tempting to keep adding tools and libraries. I’m deliberately keeping the project to plain JavaScript, Python, and JSON with no frameworks, which makes it easier to understand every part of what’s being built and to debug when something breaks.
Giving Clear Direction: Early on, vague instructions led to inconsistent results, like neighborhoods spelled differently or descriptions that sounded generic. Writing a detailed guide, including a list of words to avoid like “nestled” and “must-visit,” has made a huge difference.
Understanding What’s Being Built: Because the AI writes so much of the code, I make a conscious effort to read and understand it rather than just accept it, especially for sensitive areas like login and password handling.