Softlandia-Ltd/vision-is-all-you-need

Implements document retrieval without text chunking by converting PDF pages to images and embedding them directly with ColPali, a vision language model optimized for document understanding. The system retrieves visually similar pages from Qdrant, then grounds GPT-4o's responses with the actual document images rather than extracted text. Deploys as serverless functions on Modal with a React frontend, eliminating preprocessing overhead while preserving document layout and visual context.

405 stars. No commits in the last 6 months.