Inspiration
The purpose of BuySmart by way of a Chrome extension and FastAPI backend is to help users find the best products from shopping pages that they scan using natural-language prompts.
Inspiration: We wanted to solve the real problem of shopping in a way that does not use natural-language searches (what we are looking for is spoken or written in everyday language), and much of the time they will return searchable material. So we created BuySmart to provide users the same experience as having a smart consultant.
How It Was Made: The BuySmart Extension (MV3) scans products listed on a page and then provides you with recommendations/diagrams based on your request using the BuySmart Backend (FastAPI) which provides a deterministic ranking system, cached price history simulation, and LLM-based explanations for all recommendations.
Binomial-Probability Scoring is based upon both intrinsic and market signals and uses the following formula to calculate the overall score.
Y = 0.65Q_0 + 0.35Q_m
Challenges: Cleansing noisy scraped listing text; Supporting startup stability for users of small Railway instances; Simulating price history on listings that looks realistic; Providing fast UX while maintaining accuracy of recommendations.
What We Learned: To provide a useable experience for AI related to products there are four main areas you need to focus, prompt normalization of data, deterministic logic, providing fall-backs, and clearly explaining the recommendation given to you by your AI.
