LLM API Gateway with Observability

Teams integrating LLMs needed to manage separate SDKs and connection logic for each provider. There was also no visibility into how those LLM calls were performing across the different integrations.

Built a set of dedicated AWS Lambda functions, one per LLM provider (Gemini and OpenAI), each acting as a clean API wrapper around that provider's SDK. Any frontend or backend can freely switch between models simply by calling the relevant Lambda Fn URL, without needing to manage provider-specific SDKs or credentials directly. After the initial build, I used LLM-assisted code generation (before having access to Claude or Cursor) to plan and implement LangFuse observability into the Lambda functions adding prompt tracing, latency monitoring, and usage logging. This project was also my first real experiment with using LLMs to generate and integrate code into an existing working codebase, giving me hands-on experience with prompt engineering for code tasks and iterating with AI on real deployments.