AI & Automation
Build useful AI workflows with evidence, reproducible examples and human review.
Build an Invoice Review Workflow with AI
Use synthetic invoices to separate extraction, validation, review and payment approval.
How to Use AI for Financial Research
Build a source ledger, verify citations and recalculate figures before relying on an AI answer.
AI Memory: Decide What Your Assistant Should Keep
Keep useful facts, remove stale instructions and test what an assistant remembers.
AI Workflow Costs: Price a Completed Task
Count review, retries and mistakes alongside model charges.
Prompt Injection: Keep Documents Out of Command
Stop instructions hidden in emails or documents from becoming unauthorised actions.
AI Agent Tests: Check the Work, Not the Demo
Build a small test set that checks completed tasks, permissions, errors and repeatability.
A fluent answer is not a verified source, and extracted invoice data is not payment approval. This desk explains how to give an AI system a bounded task, inspect its output and keep consequential actions in an appropriate review process.
The first guides work through financial research and invoice handling. They show how to maintain a claim ledger, check citations, recalculate figures and test a workflow with synthetic documents whose answers are known. Proposed tests are clearly distinguished from experiments actually run, and fictional data is kept separate from real business records.
Begin with AI for financial research, then explore invoice review with AI. Continue to software portability when choosing the tools that will hold the resulting records.