Skip to content
Projects · 9 open-source systems

We turn research into running systems.

Every project is public on GitHub. The metrics on this site are the ones stated in each repository, copied as-is.

Show me work on

9 projects

Flagship

2026 · AI Safety · Lexsi Labs AI Safety & Alignment Hackathon · Track 2b

SafetyPulse

SafetyPulse checks whether a model's internal refusal mechanism is still intact during fine-tuning, using a Circuit Integrity Score (CIS) computed from 50 prompt pairs. It gates each checkpoint PASS/WARN/FAIL, repairs failing models with SafeTune, and verifies the repair with a paired audit (AuditKIT).

14.3% → 25.2% on 119 HarmBench test promptsASR rise from benign Alpaca tuning
Spearman ρ = −0.80 [−0.85, −0.71]CIS vs ASR correlation
25.2% → 12.3% (base 14.3%)SafeTune repair (ASR)
PythonPyTorchHugging Face TransformersCohere Labs Tiny Aya GlobalLlama Guard 3

Everything else