OfficerReports: The Supervisor Agent for AI-Graded, Gamified Officer Performance

OfficerReports runs a multi-tenant platform used by security guarding companies, their site supervisors, and thousands of frontline officers, spanning an admin web portal, an officer mobile app, and a backbone of more than thirty background services. This engagement added an AI-powered performance-grading agent to that platform.

OfficerReports' site supervisors manually reviewed officer-generated daily activity and incident reports across many customer sites, while separately keeping informal tabs on who showed up on time and who called out of shifts. The scale made this unmanageable: hundreds of officers per customer, dozens of customers, one to three reports per officer per shift. With effectively zero aggregated officer-level KPIs, performance reviews, coaching, and promotions all ran on gut feel and spot checks.

Taller built the Supervisor Agent: an AI-powered, gamified performance-grading system layered onto the platform's multi-tenant architecture. A hybrid grader ran deterministic checks first, then Azure OpenAI scored qualitative dimensions like clarity, completeness, and professional tone. Shift-behavior signals added punctuality, completion, and no-callout data. XP accumulated in an immutable transactions table, driving permanent rank progression, and each month produced a locked letter grade surfaced in the web portal and mobile app. Three idempotent Hangfire background jobs processed events safely on sharded SQL Server, with EF Core query filters enforcing tenant isolation. The whole pipeline was gated by each customer's opt-in to the AI suite; customers who had not opted in generated zero Supervisor Agent activity, and per-event gating snapshots preserved the audit trail. The grader shipped only after clearing a hard calibration gate against a golden dataset, and rolled out in phases from internal sites to a pilot customer to broad availability. The stack spanned ASP.NET Core 9, Semantic Kernel, Angular, and .NET MAUI, delivered by a tech lead, two to three backend engineers, a frontend and mobile engineer, and a PM.