Mike_Nobel

We developed an image-based vehicle sourcing system that enables users to identify purchase locations by uploading car photos. Our frontend (JavaScript, HTML, CSS) captures user-submitted images, which undergo multimodal binarization for feature extraction (e.g., logos, body shapes). A Python-orchestrated agent, powered by the deployed Gemma 3-4B LLM, analyzes these features to initiate targeted webscraping (Beautiful Soup/Selenium) of dealership inventories and pricing databases. The agent’s logic—developed for contextual reasoning and inventory matching—was deployed via FastAPI endpoints, which serve real-time purchase options by synthesizing scraped data, LLM insights, and image-derived attributes.