Agentic AIadvanced
Bank's Account Payables AI Agent System with Advanced AI Governance
Custodian is a governed multi-agent finance-ops platform: AI agents read invoices, score them for fraud/risk, route approvals, and auto-pay the safe ones. Six real governance layers wrap the agent runtime — Identity, Data, Model, Policy, Agent Runtime, and Operations — so every agent action is auditable and provable after the fact. It's a full Docker stack (Keycloak, SPIRE, Infisical, OpenMetadata, Presidio, LiteLLM, MLflow, Langfuse, Prometheus/Grafana) calling real LLMs (OpenAI + Groq via LiteLLM), not a mocked demo
21 lectures
What You Will Learn
How to build and orchestrate a real multi-agent AI system that makes autonomous decisions (invoice approval/payment) instead of just chatting.
How to wrap AI agents in production-grade governance — identity, data privacy, policy enforcement, and auditability — so their actions are trustworthy and provable.
How to run a full real-world microservices + observability stack (Docker, LiteLLM, MLflow, Langfuse, Prometheus/Grafana) integrating actual LLM providers, not a toy demo.
System Architecture
High-level architecture overview of the Bank's Account Payables AI Agent System with Advanced AI Governance .
What You'll Build
- A multi-agent AI system that reads invoices, scores them for fraud/risk, routes approvals, and automatically pays the safe ones.
- Six governance layers — Identity, Data, Model, Policy, Agent Runtime, Operations — that watch every agent action and can prove after the fact exactly what happened and why.
- A full production-style stack (console UI, control plane, ledger, OCR/data services) wired to real LLMs via LiteLLM, with observability through MLflow, Langfuse, Prometheus, and Grafana.
Project Instructor
Sudhanshu
4+ years exp
Premium
One Subscription. 40+ Projects. Unlimited Access.
AccessMobile & Web
Krish AI
Beta