tomasgz7/coresync

CoreSync is a system built to automate the reconciliation of corporate training attendance records that currently take up to 30 days to sort out manually. It was created as a hackathon project for the Microsoft Agents League 2026 competition and targets companies that run certification sessions across multiple physical classrooms, where each room logs check-ins and check-outs separately into different systems. The core problem it solves is that a candidate is only considered present if both a check-in and a check-out are recorded, but these two events often end up in different classroom databases with slightly different employee ID formats. On top of that, some records are corrupted or injected by HR systems in ways that can create false matches. Sorting through all of this by hand is time-consuming and error-prone. CoreSync handles this by running five AI agents in sequence. The first agent cleans up the incoming records, standardizing employee ID formats and isolating any corrupted entries. A policy connector then pulls in a set of audit rules from a Microsoft Foundry knowledge base and injects them into the next agent's instructions. The second agent, the Reconciler, applies a three-step reasoning process: it plans what to check, works through the logic, and then critiques its own conclusion before finalizing a verdict, citing the specific audit rule behind each decision. The third agent classifies the results into three groups: present, absent, or unresolved. The fourth agent handles follow-up for absent employees, choosing the right time and channel to reach them based on signals from their Microsoft 365 calendar. The fifth agent produces anonymized summary reports for managers. The system is built in Python and runs on Microsoft Azure infrastructure. It is intended to replace a manual process with one that produces real-time results and keeps a full audit trail showing exactly why each attendance decision was made. The full README is longer than what was shown.