Developed and delivered a new feature for the jeejeelee/vllm repository, enabling vLLM Chat Completions to accept prompt_embeds content parts. This addition allows users to send pre-computed embeddings alongside text in chat messages, increasing flexibility for embedding-driven prompts and improving performance. The work involved designing and updating the Python API, expanding documentation, and providing example scripts to demonstrate usage. Comprehensive tests were implemented to cover integration and edge cases, and the end-to-end workflow was validated within the CI pipeline. The project showcased skills in API development, deep learning, and Python programming, with a focus on maintainability and usability.