UtilityOps Readiness

UtilityOps was created to test an idea about what AI could contribute in an operational environment. It does not come from professional utility-industry experience and is not presented as a deployed utility platform.

I defined the problem, data flow, synthetic work-order examples, and test cases. AI assistance contributed substantially to the technical architecture and code. The application examines indexed records, applies deterministic checks where known rules can answer the question, and uses retrieved information with a language model where notes or document context require interpretation.

The demonstration used external model services because the data was synthetic and stronger indexing was needed than the local models available to me could provide. Testing included three distinct work-order cases, including one with a less obvious blocker embedded in the source documents that the system needed to identify.

The project helped me work with retrieval, embeddings, structured answers, citations, deterministic validation, context limits, and hallucination risk. Prompt changes, repeated extraction tests, deterministic loops where possible, and removal of unnecessary context were used to improve output behaviour.

UtilityOps is available in a public GitHub repository with instructions for running it. It is currently paused. A meaningful next step would be a more deliberate evaluation set that measures whether blockers, supporting sources, and readiness decisions are identified consistently.