Project · 2026
Casefile
AI investigation assistant
Drafts anti-money-laundering case write-ups where plain code, not the AI, checks every figure.
- 417 / 417
- deliberately corrupted claims flagged
- 0 / 994
- clean claims falsely flagged
- 36
- automated tests in CI
PythonLangGraphFastAPISQLiteReactTypeScriptscikit-learnDockerGitHub Actions
Limits: Synthetic ledger. The AI writing step has only run on a stand-in model so far; real-model testing is next.
Read the case study →