Compliance Dashboard for AI Teams

I built this dashboard for a team that ran eleven AI agents across support, sales, and scheduling. Each agent logged in its own format. Release reviews took two days of searching log files and asking developers what each tool call did.

The dashboard gathers the whole fleet in a single view. It streams tool calls from every agent into one audit table with session id, tenant, tool name, and policy decision. A lineage view traces each answer back to source documents and retrieval calls. Auto-flagging marks sessions with unapproved outbound messages, raw PII in logs, and retention stamps past due. Reviewers open flagged sessions first and clear the rest in bulk.

Export takes one click. It produces a dated report with call counts per agent, approval coverage, and deletion counts from the nightly job. The security lead signs releases from this report, and auditors read the same pages during reviews.

Review time dropped from two days to twenty minutes. The team ships weekly now, and the dashboard catches two or three policy misses per month before customers see them. I repeat the same twenty-minute review at the start of every month.