I presented Tracing an LLM API Request as part of Major League Hacking’s Hacktoberfest 2026 livestream series.

In this visual talk, we follow a single LLM API request from the client to the GPU and back, exploring the inference pipeline, performance bottlenecks, and optimizations that make modern LLM serving possible.

Visual Explainer (5 min)

A short, animated journey through the LLM inference pipeline, from API gateway to GPU execution and streamed response.

Along the way, see how techniques like KV caching, cache-aware routing, continuous batching, and speculative decoding improve performance.

Watch the Full Talk (45 min)

Watch the full Hacktoberfest livestream for a deeper walkthrough of the inference pipeline, practical examples, and Q&A.

Resources

Hacktoberfest livestream event details for Tracing an LLM API Request
Sujee Maniyam presenting Tracing an LLM API Request during the MLH livestream
Speculative decoding slide from the Hacktoberfest talk
Q&A and thank-you slide from the Hacktoberfest talk