Capacity is a support automation platform whose Answer Engine helps enterprises surface answers from across fragmented knowledge sources — documents, emails, chat threads, PDFs, videos, and apps. Traditional search lacked the context to bridge these information silos, so the company set out to move from a natural language classifier to a generative AI-native architecture that could deliver precise, context-aware answers at enterprise scale.
Working with Microsoft, Capacity deployed Phi-4 and Phi-4-mini small language models via Azure AI Foundry for their speed, cost-efficiency, and flexible deployment options. The modular architecture splits Phi tasks between offline preprocessing (title generation, keyword tagging) and real-time query refinement, supported by Azure AI Translator across 46 languages, Azure AI Language, Azure AI Search for semantic retrieval, Azure Database for PostgreSQL, Azure Kubernetes Service for containerized deployment, and Azure AI Video Indexer for multimedia content.
The rearchitected Answer Engine delivered 4.2x cost savings versus the prior pipeline, 97% first-shot tagging success, a 56% accuracy improvement, and reduced summarization time from 12–14 seconds to 4–5 seconds. Capacity is now exploring Phi-4-multimodal for complex reasoning and image understanding, with further plans to enhance knowledge graphs and interoperability across institutional knowledge bases.