Aegis-riskguard

The problem Inside a regulated bank, the 2nd Line of Defence (2LoD) Operational Risk team is responsible for:

Maintaining a coherent set of risk policies (Operational, Strategic, Reputation, ERM, Risk Appetite) Aggregating RCSA (Risk & Control Self-Assessment) submissions from every business unit Detecting gaps between what policy requires and what departmental controls actually do Drafting remediation memos and tracking issues to closure Today, this is done with Excel, Word, email, and a lot of senior-analyst time. Reviewing a single departmental RCSA against five policies takes hours. Consolidating 16 RCSAs across a bank, weeks. The work is repetitive, time-pressured, and high-stakes — exactly the wrong combination for human-only execution.

Aegis-RiskGuard automates the analytical grunt work while keeping a human firmly in the loop for every decision that matters.

Why not just "use ChatGPT"? Three reasons. They're the entire reason this project exists:

Prompt injection in untrusted documents — A policy PDF or RCSA spreadsheet uploaded by a malicious insider can pivot a naive copilot. Aegis blocks this at the firewall. No audit trail — Regulators reject "the model said so." Every prompt, response, and policy decision in Aegis is logged with an HMAC-signed entry. No declared-vs-detected intent enforcement — You don't know what your agent thought it was doing. Aegis catches mismatches between agent declarations and content. These are not features. They are the minimum for any AI system that touches a regulated 2LoD function. DORA Article 5–14 and EU AI Act Annex III say so.

Note: the demo URL is substituted with the GitHub repository link.