Course content.
5 weeks (self-paced) · 40 hours · 5 lessons
Foundations of LLM-Assisted Code Review
How transformer models process Solidity — strengths and blind spots · Prompt engineering patterns for security review (few-shot, chain-of-thought) · Setting up a reproducible AI-audit workbench
Retrieval-Augmented Audit Pipelines
Chunking, embedding, and indexing large contract codebases · Feeding context windows with the right files at the right time · Structured output: getting findings as severity/location/remediation JSON
Toolchain Integration
Piping Slither and Foundry output into an LLM for triage · Building custom MCP tools for audit-specific queries · Automating first-pass review with CI-triggered model runs
Hallucination Detection & Quality Control
Common hallucination patterns in code-review models · Cross-checking model output against on-chain state and test results · Building a human-in-the-loop review gate
Capstone & Tool Evaluation
Full AI-augmented audit of a non-trivial protocol · Speed and accuracy comparison: manual vs AI-assisted · Evaluating emerging AI-audit products with a structured rubric