SASHA

I built Sasha, a multimodal HS Code classification agent that uses Gemini 3 to identify the correct HS Code of any given product. It helps customs and trade experts perform custom clearance faster and accurately.

Gemini 3’s enhanced reasoning capabilities let Sasha perform knowledge based deep reasoning, while perfectly handling a multi-step classification process, which allows customs and trade experts to get accurate classification. That saves time and money.

SASHA also leverages Gemini 3’s multimodal architecture, enabling users to choose betwenn inputing a text description or uploading an image/list of the product they want to classify, through a simple modern streamlite chat interface.

While the system integrates two third-party services, Gemini 3 remains the central reasoning and decision making layer by seamlessly integrating in backend workflows which were implemented using langflow and deciding how and when to use the chroma knowledge base.

By placing Gemini 3 at the core of SASHA, the application delivers a unique, easy to use, and intelligent solution that seamlessly combines multimodal input, state of the art AI-driven reasoning, and world custom organization data, to bring real value today while remaining flexible for future globalization.

Built With

• chromadb
• langflow
• render
• streamlite