I need a compact microservice that exposes a single REST endpoint where my PHP-based application can POST an item name. The service should forward that string to a pre-trained model running on Google Cloud (Vertex AI or another managed option of my choice, selectable through a config value or ENV variable). Inside the microservice you will: • Pull embeddings or similarity scores from the chosen model, comparing the request against my catalogue of roughly 500 products. • Rank results primarily by item keywords, then package the best one (or top few if the score gap is small) into a JSON payload that contains: the catalogue item name, a numeric confidence score, and a short natural-language justification explaining why the match was chosen. Tech expectations – Written in PHP so it slots straight into our stack; Slim, Lumen, or plain Laravel routing is fine as long as it stays lightweight. – Clean, well-documented code and a README covering local run, GCP credentials, and model swap-out steps. – Response times suitable for realtime look-ups (think <500 ms once warmed). Acceptance criteria 1. `POST /match` returns HTTP 200 with `{"item":"…","score":0.92,"reason":"…"}` and optional array of alternates. 2. Switching models requires only an ENV change and a redeploy, with no code edits. 3. Unit test set includes at least 5 sample queries proving correct ranking behaviour. If you have previous experience wiring PHP services to Vertex AI or similar, let me know. I’m ready to start as soon as we nail down your approach.