Although we are inundated with articles and tutorials promising GenAI-based chatbots in just five minutes or so, the reality of developing a reliable agent is much more challenging. Some predict that 30% of projects will be abandoned due to the difficulty of translating proofs of concept and substantial investments into tangible benefits. To help prevent this, an open-source rapid prototyping platform called "Oracle AI Microservices Sandbox" has been released. This platform enables zero-code development of knowledge bases using Oracle DB 23c AI and its Similarity Search capabilities, allowing users to select any LLM for vector embedding and chat completion, whether on-premises or in the cloud, starting from template-based prompt engineering. The hyperparameters related to the LLM and the knowledge base processing can be tested either individually or in bulk, facilitated by the generation of Q&A test datasets to determine accuracy rates. Successfully tested configurations, along with the generated knowledge base, can be published and exported as AI microservices using Spring AI or LangChain, accessible through OpenAI API-compatible REST endpoints.

Accepted @ Analytics and Data Summit 2025