I build the parts of software that don't have a Stack Overflow answer.
Javair Ratliff — an independent software and AI engineer in New York. Nine years spent carrying hard technical work, from language-model internals to computer vision, out of the research notebook and into dependable production.
Ideas from the lab, drawn until they hold under load.
Engineering across the full stack of modern AI.
AI & Language Models
Representation engineering, model-behavior modification, and inference pipelines — taking a method from a paper to a system you can actually deploy. Creator of the abliteration technique.
Full-Stack & Cloud
Serverless architectures, APIs, and backends on AWS — built to scale cleanly from prototype to thousands of daily users without a rewrite waiting on the other side.
Computer Vision & ML Systems
Real-time, GPU-accelerated vision and pose-detection systems, plus the unglamorous deployment plumbing that keeps models running reliably in the wild.
Prototyping & Technical Advisory
Credible prototypes for hard technical bets, and the architecture review that tells you — honestly — whether an idea survives contact with scale.
A few things I've built.
Abliteration↗
An open-source representation-engineering technique for steering large language model behavior through direct weight editing — no fine-tuning required. Adopted widely across the open-source AI community.
Real-Time Vision Exhibit↗
3D ball-trajectory and speed tracking for an interactive exhibit at a New York museum, built on custom GPU-optimized algorithms and seen by thousands of visitors every day.
Serverless Health Platform↗
An AWS Lambda and DynamoDB screening platform that scaled to thousands of daily users during COVID-19 reopening, driven by automated SMS compliance workflows.
Nine years from idea to production.
I've spent nine-plus years building production systems — from GPU-optimized inference pipelines to serverless platforms serving thousands of people a day. My work sits where machine-learning infrastructure, language-model internals, computer vision, and distributed systems overlap.
I'm at my best with teams who have a genuinely hard problem and need someone who can move between research and shipping without dropping either. Self-taught, practicing since 2008, based in New York and working with teams anywhere.