As we continue to navigate the AI landscape with new and exciting breakthroughs, the cost of deploying such novel solutions has seen a rapid and exponential escalation. Current LLM models require millions of dollars to train and deploy in software, hardware, and skilled manpower costs. Coupled with this is the problem of training data confidentiality which enterprises are reluctant to let go from their premises, and the security and ethical considerations of the AI results provided by such systems. If we have to generate value for enterprises using AI tools, it is imperative that we reduce the cost of such deployed systems, figure out ways of deploying them on premises without prohibitively huge hardware costs, and create some guarantees regarding the security and the ethicality of the obtained results. To tackle such issues enterprises are turning to solution providers called intelligence integrators that aim to integrate the AI evolution with their customers’ evolution. This talk will touch upon the various efforts that we are undertaking at Fujitsu Research to tackle the above problems. It will elaborate on our research on creating leaner and customized AI models and in making training less expensive. To enable on premise AI deployment, we have created efficient knowledge graph-based RAG systems and domain specific LLMs. I will describe such efforts and also our work on creating ethical and security guarantees of our AI generated results. The talk will showcase a number of real-world solutions that are already in deployment with our enterprise customers and demonstrate how such solutions are currently generating value for them.