Developed and delivered a multimodal inference endpoint for the jeejeelee/vllm repository, enabling the processing of both text and image inputs within a unified API. Focused on backend and API development using Python, the work introduced new encoding and decoding utilities to streamline the interaction between rendering and generation phases, supporting complex multimodal requests. The implementation enhanced the /inference/v1/generate endpoint with robust multimodal support, aligning frontend and backend workflows. Emphasis was placed on clean integration, scalability, and maintainability, with thorough testing and collaborative code governance. No major bugs were reported, reflecting a focus on delivering business-valued, production-ready features.