wtf is locallysells-voice-search?

This project is a voice search and voice ordering assistant built for LocallySells, an online marketplace that delivers tobacco and nicotine products locally. It was created during a hackathon called the Multimodal Midsummer Hackathon, and it lets a caller phone in, describe what they want in plain conversational language, and have the system find matching products and place an order. The core idea is understanding meaning rather than exact keywords. If someone says they want something minty and refreshing, the system can match that description to a specific product like a mint flavored vape, even if the caller never says the brand name. It does this using semantic search: each product in a 198 item catalog has been converted into a numerical representation ahead of time, and when a caller speaks, their words are converted the same way so the system can compare and rank the closest matches. A keyword check is layered on top so that when someone names an exact brand, that brand still comes out on top. Three separate services work together to make a phone call happen: one handles the live voice conversation, turning speech into text and back into speech, a second acts as the reasoning brain that decides whether the caller wants to search for a product or place an order, and a third hosts the actual search and ordering logic as a small serverless function, along with the product catalog and its embeddings. The system includes safety rules: it will not serve anyone who indicates they are under 21, it will not invent products or prices that are not in the catalog, and it avoids giving medical advice. The project also includes automated test suites that simulate callers with different speaking styles, checking that the assistant handles vague requests, misremembered product names, and multi step orders correctly, rather than relying on manual spot checks. The repository includes Python scripts for building and deploying the function, precomputing embeddings, and setting up the voice assistant, along with documentation describing how a call flows through the system. There is no license mentioned in the README.