security engineer
into math, multimodal ai, and figuring out how things break.
security engineer
into math, multimodal ai, and figuring out how things break.
Agent harness for authorized bug bounty research with safety tiers, autonomous campaigns, reusable security skills, and evidence-first reporting
Python 1
Does medical-reasoning fine-tuning make a small LLM safer or confidently wrong? A controlled before/after calibration study on Qwen3-4B + MedGemma-4B (single free Kaggle T4). Finding: internal cali...
Python
Browser extension + local proxy that strips PII from LLM requests before they leave your machine. Presidio + spaCy NER, all detection runs locally.
JavaScript
Clinical tools for DeepSeek Harness: handover and discharge drafts, de-identification, completeness checks, and cited lookups. Draft-only, not a medical device.
TypeScript 3