Applied AI Stories: Solving Cold Cases and Powering Product Search

Good vibes, and a whole lot of machine learning magic packed into one evening at ML6. From AI agents digging through cold case archives to smarter search on IKEA.com! 🧠✨

??️ Talk 1 - Can AI agents help solve Europe’s Largest Cold Case?
Titus Naber kicked things off with a deep dive into the Olof Palme investigation - one of Europe’s biggest unsolved mysteries. His Deep Research agent system showed how multiple AI agents can collaborate to sift through mountains of police reports, cross-reference clues, and even generate new hypotheses.

What to think about when designing collaborative AI agents?
🔸Provide multiple “search angle” tools to your LLMs (filtering options, access full documents vs. chunks, search per type, time, location etc.).
🔸Aligning with the investigating journalists was key to get domain research procedures into the agentic system - so get your domain knowledge in.
🔸Think about agent specialities thoroughly upfront; a well thought design gives the best transparency and insights about the thought process.

🛍 Talk 2 - From Keywords to Concepts: A Late Interaction Approach to Semantic Product Search on IKEA.com
Then Amritpal Singh Gill from IKEA took us from crime scenes to customer journeys. He shared how IKEA shifted from simple keyword search to semantic understanding - so users find what they actually mean, not just what they type. Using late-interaction models, token-level scoring, and clever training tricks, they built a multilingual search experience that’s fast, smart, and truly helpful.

That search is no longer about words, that’s clear. But what else?
🔸The Multi vector (late interaction) approach kept token granularity and importance, and smaller embedding sizes meant lower latency.
🔸Synthetic product descriptions and search queries were used to unlock speed and coverage, and in the end create a custom and accurate ColBERT model.

Huge thanks to ML6 for hosting, to our brilliant speakers Titus Naber and Amritpal Singh Gill, and to everyone who joined us tonight!