Microsoft Ignite 2025: Semantic Caching Breakout & Product Booth & DP-900

Location: San Francisco, CA (Microsoft Ignite)
Role: Breakout session speaker; booth duty product expert
Audience Size: Conference breakout session (50 attendees) & 1:1 booth conversations

I served as a speaker at Microsoft Ignite 2025, delivering a breakout session where I demonstrated a production-minded semantic caching solution built to reduce LLM cost and latency while improving user experience. The talk focused on practical architecture: embedding-based similarity lookup, Redis-backed cache retrieval, and real-time app behavior that still preserves the economics of reuse.

In addition to the session, I supported customers and partners directly on the show floor as a product expert, answering implementation questions, discussing sizing and operational considerations, and translating architecture into deployment-ready guidance. I also used the event to take and pass the data fundamentals exam, DP-900 certification, onsite.

(Topics covered in the talk):

• When semantic similarity is “good enough”: picking thresholds, handling near-misses, and avoiding incorrect reuse.

• Cache freshness and correctness: TTL strategy, invalidation patterns, and what consistency guarantees you can (and can’t) claim.

• Measuring ROI beyond vibes: how to model savings from reduced LLM calls and what metrics to track in production.

• Designing the cache key space and stored payloads: what to store (prompt, response, metadata, embeddings) so the cache is reusable and debuggable.

• Operational readiness: sizing for vector workloads, monitoring hit rate/latency, and preventing cache growth from becoming a silent tax.

• Delivered a breakout session demonstrating an end-to-end semantic caching experience and implementation approach.

• Supported real-world architecture and deployment conversations at the booth, refining positioning and common solution patterns.

• Passed DP-900 onsite, reinforcing data platform fundamentals for data and cache discussions.