SenPy

The challenge

Users needed to ask questions that mixed text and images in one conversation — describing a photo, asking follow-up questions, and getting responses that stayed grounded in what was actually shown, not just the text prompt.

Our approach

• 1

  Built a multimodal input pipeline handling text and image uploads in the same conversation thread.

• 2

  Integrated OpenAI's vision-capable models with a conversation memory layer so image context carries across follow-up questions.

• 3

  Designed a streaming response UI so answers appear token-by-token instead of a long wait.