AdAudit is a web-based enterprise AI agent for paid media teams. Instead of acting like a normal ad copilot that simply generates campaign ideas, AdAudit behaves like a guarded media buyer: it studies the product brief, audience, budget, landing page, and creative asset, then decides whether a campaign is safe enough to prepare. The agent collects evidence, applies paid-media playbooks, uses Gemini on Vertex AI for multimodal creative review, compares multiple Meta-style launch strategies, evaluates budget economics and delivery readiness, and then repairs unsafe plans before execution. If a creative contains risky claims such as guaranteed employment or unrealistic outcomes, AdAudit rewrites the claim into a safer proof-first angle. If the budget is too thin for too many ad sets, it consolidates the structure. If the pixel or conversion signal is not ready, it changes the objective to a safer first-flight signal. The key safety boundary is that AdAudit never creates an active campaign. Its executor only prepares Meta-compatible PAUSED campaign objects, with human approval required before any real spend. The system includes program-level causal guardrails, so safety checks are not just LLM self-reports. The app is deployed on a Vultr VM, served as a full-stack Node and React application, and integrates Google Gemini through Vertex AI Application Default Credentials. AdAudit targets real enterprise friction: AI agents can now plan and operate ad workflows, but businesses need a trustworthy layer that can research, reason, repair, and stop before money is spent. It demonstrates agentic workflows, enterprise utility, multimodal intelligence, and a production-style deployment.
Category tags: