For billions of people, digital communities hold a goldmine of authentic advice—from car repairs to gardening—but finding it often feels like searching for a needle in a haystack. This presentation details how we are using Large Language Models (LLMs) to eliminate the friction between users and the expertise they seek. We address two core frustrations: the failure of traditional search to understand natural language intent (like missing "cupcakes" when searching for "small cakes") and the fatigue of scrolling through endless comments to find a consensus. By blending "concept matching" with keyword search and deploying AI to summarize complex threads into instant insights, we help users find answers, not just links. We validate this improved experience using a novel "LLM-as-a-Judge" framework to ensure results are genuinely helpful, transforming platforms into intelligent engines of collective wisdom.

https://arxiv.org/abs/2509.13603