Token Efficiency & Air-Gapped MLOps

All right!!!

Meetup #36: "Token effciency & Air-gapped MLOps" with Cloudera, Qdrant, evroc and SAAB will take place June 3 at AI Sweden. The program is now final! :-)

Event Program
- Doors open at 17:00 CET
- Talks begin at 17:45 CET
- There will be pizzas & drinks
- There may be a moderated Q&A session....

Speaker Line-Up
- Ewa Szyszka, DevRel Engineer at **Qdrant **will give a talk titled "Grow the Data, Shrink the Bill". Abstract: Agent efficiency through vector search. The core argument: AI agents waste tokens in four ways: context accumulation (stale tools, redundant reads), payload inefficiency (loading too much context), generative waste (restating, re-planning), and failure recovery (bad tool calls, self-correction loops). Stuffing the context window with everything works but is 50–100× more expensive than targeted retrieval. The alternative: use vector embeddings to increase information density, so an orchestrator works only with retrieved context from a vector store. You back this up with a benchmark experiment across three datasets (SWE-Bench, LongMemEval, FinanceBench) comparing agentic grep, Qdrant BM25 sparse, and Qdrant dense retrieval, measuring tokens-to-resolution as the primary metric.

- Markus Kjellner, Regional Vice President, Nordics & Baltics at Cloudera will give a talk titled "GenAI and MLOps without compromising on data sovereignty in strict air-gapped security environments". Abstract: Unlock the power of GenAI and MLOps without compromising on data sovereignty or strict air-gapped security. Discover how you can bring cutting-edge, cloud-native AI capabilities directly to your data within fully disconnected and sovereign environments.

- Peter Sundström, AI & Digitalisation Transformation Lead at SAAB will give a talk titled "MLOps for Electronic Warfare". Abstract: MLOps for electronic warfare enables rapid deployment and updating of ML models for RF signal detection and threat classification. Models often run on edge platforms with limited connectivity and strict latency requirements. The main challenge is keeping models reliable in dynamic and adversarial spectrum environments.

- Mikael Vesavuori, Product Manager AI Products, evroc will give a talk titled "AI Serving Across the Control Spectrum". Abstract: AI serving is shifting from simple cloud APIs toward a broader spectrum of deployment models: sovereign cloud, dedicated environments, on-prem, hybrid, and air-gapped AI." In this talk, Mikael Vesavuori from evroc explores when different AI serving models make sense, what cloud enables compared to isolated environments, and why the future may be less about choosing one model and more about controlling where each AI workload runs.

Looking forward to seeing you there,

/Patrick & the Stockholm MLOps Team