In medium-sized manufacturing companies, preparing quotations for custom parts is often time consuming and heavily reliant on expert knowledge. We present CoLIBRi (Conversational Image-Based Retrieval), an end-to-end system for multi-modal retrieval and conversational search on image-based manufacturing drawings. The system supports quotation by retrieving relevant historical drawings together with associated manufacturing data. Our approach combines fine-tuned optical character recognition for textual attributes (e.g., dimensions, tolerances, materials), deep-learning-based segmentation for part geometries, pre-trained visual embeddings for geometric representation, and a vision-language model to extract structured metadata such as part names from title blocks. Engineers can either upload a drawing to trigger multi-modal retrieval or use a conversational interface powered by a large language model to issue natural-language queries and explore retrieved parts. By linking retrieved manufacturing drawings with enterprise resource planning data on machine usage and production times, CoLIBRi enables faster quotation preparation while reducing dependence on individual expertise. Developed in close collaboration with a medium-sized manufacturing company, the system demonstrates how multi-modal AI and conversational interfaces can provide practically relevant support for industrial engineering workflows. To foster reproducibility and future research, we release the CoLIBRi model weights and source code under an open license.