Talks, workshops, events, and training for AI Engineers.
Stop waiting for Docker builds to finish just to test a single model tweak. RunPod's Flash SDK lets you deploy GPU-accelerated functions directly from your IDE using a simple Python decorator, slashing iteration time from minutes to seconds.
While social media claims RAG is obsolete, Cursor’s latest data shows a 24% spike in model accuracy when using semantic indexing. The secret lies in treating embeddings as cached compute, allowing agents to bypass the expensive token-burn of raw file grepping to find the exact million tokens that actually matter.
The boundary between research and production is dissolving as San Francisco prepares for a massive convergence of engineering talent. While the industry talks about cooling systems, the real pressure is in the 'heat' of shipping reliable AI at startup speeds.
The 'eval' function is no longer a security nightmare. Cloudflare is introducing Dynamic Workers to safely execute LLM-generated code in locked-down isolates, effectively reclaiming three decades of lost compute potential while collapsing thousands of API endpoints into tiny token footprints.
Your AI agent just called tool B before tool A, ignoring a critical dependency, and your code didn't catch it. Dat Ngo reveals why traditional auditing is dead in the age of non-deterministic systems and how telemetry is the only way to track what your models are actually doing. There is a way to automate yourself out of the observability loop entirely.