"Code2Create team on IBM Bob 2.0 hackathon Hackathon"

Team Idea

Axiom: AI tools like IBM Bob make writing code fast, but they introduce a dangerous blind spot: unwritten assumptions. When an AI agent generates a multi-file PR, it silently assumes edge cases won't happen—that payloads are never empty, external APIs won't timeout, or database writes are always atomic. Because the generated code compiles cleanly and passes basic happy-path tests, developers end up rubber-stamping changes they haven't thoroughly stress-tested. Axiom acts as an adversarial auditor for AI-generated code. When a PR is created, Axiom scans the diff, uncovers the hidden assumptions the AI made, and automatically generates targeted edge-case tests to deliberately break them. If a test fails, Axiom isolates the issue, maps the blast radius across downstream services, and sends the reproduction steps directly back to the AI dev partner to auto-patch before merging to main. It turns code review into automated, zero-trust verification.

Team Members

Anmol Suwalka

Anmol_Suwalka

Submission

Axiom: Invariant Firewall for AI Devs

Axiom audits code generated by AI dev partners like IBM Bob, uncovers hidden unwritten assumptions, executes synthesized adversarial stress tests to break them, and maps downstream blast radius before merging to production.