Skip to content
AI|Course

Designing system architectures · Beginner

Architect Agentic Systems for Financial Workflows

Design systems where language models help with financial workflows without being trusted with what they should not decide: exact money handling, validated extraction, approvals, audit trails, evaluation and safe rollout. Built on synthetic data only.

7 lessons · 15 hr 15 min · Free

You will learn to

  • ▸Explain why finance workflows demand determinism, auditability and controls
  • ▸Design a pipeline with the model at the edges and rules, approvals and limits at the core
  • ▸Handle money exactly, and have models calculate through tools instead of in their heads
  • ▸Extract and verify data from documents with evidence and validation
  • ▸Evaluate, roll out and monitor an automated workflow, and map it to risk guidance

Before you start

  • ·Comfortable with Python or TypeScript
  • ·You have called an LLM API with tools or structured output
  • ·No finance background needed; the course explains each workflow

Syllabus

  1. 01Why finance is differentWorkflows, stakes and the principle: the model assists, the system decides.45 min
  2. 02Architecture patterns for financial workflowsThe reference pipeline, case state machines and exception queues.1 hr 30 min
  3. 03Numbers, data and ledgersExact decimals, calculation tools, authoritative sources, double entry.2 hr
  4. 04Documents, extraction and retrievalSchemas with evidence, validation, and what FinanceBench showed.2 hr
  5. 05Controls, compliance and risk guidanceApprovals, audit trails, model risk management and frameworks.2 hr
  6. 06Evaluation and operationsError-cost metrics, shadow mode, monitoring and incidents.2 hr
  7. 07Capstone projectBuild a controlled financial workflow on synthetic data.5 hr