Sqwish AI

Sqwish AI, founded in 2024 by Ushnish Sengupta and Federica Freddi, is a Cambridge-based company developing an efficiency and optimization layer for artificial intelligence applications. The founders, who are second-time co-founders, leverage their deep backgrounds in AI research from the University of Cambridge. Ushnish Sengupta holds a PhD in Engineering, specializing in Bayesian machine learning and optimization, with prior experience at Mediatek Research and Rolls-Royce. Federica Freddi holds a Master of Engineering from Cambridge and has a background in AI, software development, and deep learning from her time at Mediatek Research and ARM. Their previous venture together, IDEOM AI, focused on LLM-powered creative tools, laying the groundwork for their current focus on AI efficiency.

The company's core technology is designed for developers building applications on top of large language models (LLMs). Sqwish AI's platform uses real-time reinforcement learning to continuously optimize AI applications based on specific business key performance indicators (KPIs), such as conversion rates and user retention, alongside technical metrics like cost, latency, and accuracy. This is delivered through an easy-to-integrate software development kit (SDK) that acts as a performance flywheel; each user interaction enhances the intelligence and effectiveness of the entire AI stack. The platform is compatible with major models including those from OpenAI, Google Gemini, Claude AI, and Meta Llama.

Sqwish AI's product addresses the significant operational costs and latency challenges in generative AI. It offers a suite of optimization tools, starting with prompt compression, which can reduce the size of inputs sent to LLMs by up to 10x without sacrificing the quality of the response. This capability is particularly beneficial for applications involving retrieval-augmented generation (RAG), document processing, and complex agentic workflows. Beyond prompt compression, the platform provides adaptive model routing to select the most cost-effective model for a given task, real-time prompt and context adaptation, and personalization based on user behavior. The business model enables AI software developers to improve operational efficiency, increase application accuracy, and reduce resource costs, ultimately allowing them to build more capable applications with less expenditure.

Keywords: AI optimization, prompt compression, large language models, LLM efficiency, reinforcement learning, AI cost reduction, generative AI, RAG optimization, model routing, AI performance management, context optimization, adaptive AI, AI SDK, natural language processing, Ushnish Sengupta, Federica Freddi, AI application development, latency reduction, business KPIs for AI, AI stack intelligence