Скачать или смотреть Shop Assistant - OpenAI Open Model Hackathon Submission using GPTOSS-20b

This is inventory management made human—just talk, and it works.

Shop-Assistant is a smart voice-powered inventory companion designed for retailers. It enables both customers and staff to manage stock simply by speaking. Users can add or update items verbally, and the system handles transcription, structured understanding, and idempotent inventory updates seamlessly.

The workflow is as follows: mobile clients upload audio (or stream it live) → Google Cloud Speech-to-Text produces a transcript → an NLU service calls the gptoss-20b model with a strict prompt that returns validated JSON (items, quantity, unit, intent) → the Inventory service applies idempotent updates (via request_id) to PostgreSQL and emits events to RabbitMQ. Authentication is handled via JWT.

With Shop-Assistant, managing stock is effortless, inclusive, and reliable. Retailers save time, reduce errors, and keep shelves stocked, even in areas with poor connectivity.

We created this project during this hackathon and integrated gpt-oss-20b on Replicate to transform raw speech into structured, actionable inventory updates.