Featured Project
Project Helix — Handshake AI
Role: AI training data engineer at Handshake AI
Contributed to an AI model training pipeline by authoring high-quality, deterministic
fail-to-pass test suites across real open-source codebases — a SWE-bench–style
evaluation framework used to train and benchmark coding AI models. Each task required
deeply understanding an existing codebase, identifying a concrete bug or missing
feature, implementing the fix, and writing tests that fail on the original code but
pass on the solution. Each task was packaged with a Docker-based evaluation runner and signal manifest for
automated CI validation.
// fail-to-pass test: auth security
def test_verify_bcrypt_password():
handler.accounts = {"user": bcrypt_hash}
assert handler.verify_password("user", "secret")
assert not handler.verify_password("user", "wrong")
// coverage: security · protocol · db · api
// langs: Python · PHP · Rust
// eval: Docker + signal.json ✓