Worked on the livekit/agents repository to deliver stateless LLM conversation capabilities, focusing on scalable API usage and improved reliability under concurrent workloads. Refactored the LLM class in Python to eliminate stateful conversation management, removing unnecessary instance variables and simplifying the codebase. This approach enabled each API call to operate independently, reducing the risk of state leakage and enhancing concurrency safety. Emphasized asynchronous programming and backend development practices to boost throughput and maintainability. The changes addressed potential bottlenecks in testing and deployment, resulting in a more robust and scalable API integration suitable for high-demand, concurrent environments.