Each workflow follows the same pattern: the LLM layer interprets the question, the Energy Domain AI layer runs the relevant analysis against audited records, and the answer is returned with traceable evidence. Nothing in these flows touches equipment.
Site Copilot routes to the Revenue Gap Analyzer, which compares the approved dispatch plan against dispatch results, meter data, and derate records. Answer: ranked causes (e.g., a 2-hour BMS thermal derate during the peak window) with monetary impact and links to the underlying events.
The Dispatch Recommendation Engine builds a day-ahead plan from tariff, load forecast, weather, and SOC constraints. The plan is presented with all constraint checks; the operator approves it, and the deterministic EMS scheduler executes it through safety gates.
The Alarm & Fault Explanation Engine classifies the alarm, explains the cause and the affected functions, and recommends a response — for example, distinguishing an informational comm-retry from a trip-path-relevant relay alarm.
The Copilot evaluates SOC envelope, BMS permissives, thermal headroom, power limits, and any active lockouts, then answers with a clear yes/no plus the binding constraint. The actual discharge still executes only through the EMS within those limits.
The Revenue Gap Analyzer attributes the gap across derates, forecast error, missed windows, and availability loss — each line item traceable to logged events.
The Copilot reconstructs the timeline from relay signals, breaker feedback, EMS state transitions, and PCS responses — including the trip (relay-led) and the supervised, authorized recovery sequence — into a clear narrative for owner and utility communication.
The Operator Report Generator compiles verified meter data, dispatch results, savings, availability, and notable events into a formatted report. Figures cite audited records; the narrative is reviewed before sending.