Summer Home Recommender

What I Built

A recommendation system that uses LLMs to extract intent from natural language queries, then scores properties across multiple dimensions: environment, budget, capacity, amenities, and location.

How It Works

Instead of forcing users to fill out forms with checkboxes, they just describe what they want in natural language. The LLM extracts their preferences and priorities. The system then dynamically weights each scoring dimension based on what the user actually cares about.

Built both a CLI and Streamlit UI. Handles 100+ properties with CSV export for further analysis.

What I Learned

Natural language interfaces are harder than they seem. The challenge isn't the NLP part, it's handling ambiguity and conflicting preferences. When someone says they want "quiet but close to town," the system has to make tradeoffs, and those tradeoffs need to match user expectations.