eeddiedev

 I'm an AI/ML engineer who builds tooling most engineers avoid zero-dependency agent frameworks, keyless LLM eval harnesses that run on CPU, prompt-injection scanners, and smart-contract audit copilots with benchmarked recall. My work is measurable: 27/29 recall on a prompt-injection golden corpus, dense search beating lexical 94% vs 75%, an audit copilot scoring 0.97 on 1,748 real contracts. 172 public repos, shipped daily. I don't need API keys, GPUs, or excuses — I need a hard problem.