DeepSeek-ing a Needle in a Haystack

With over 1200 sessions at NVIDIA’s GTC conference, selecting the most relevant talks can feel like searching for a needle in a haystack. Traditional search methods often fall short, returning a flood of keyword matches and marketing fluff. In this talk, I’ll share how we tackled this challenge by leveraging agentic workflows with DeepSeek R1 to filter and rank sessions efficiently.

I’ll talk about how I crafted the prompts in detail, why temporal was technically necessary to solve this problem (and others). I’ll also discuss how other models could be used in stead of DeepSeek-R1 (like Gemini, OpenAI 01 etc).