Keaton Z.

AI Developer | Automation & Predictive Analytics I build AI systems that take repetitive work off your plate and turn your data into decisions you can act on. Over 5+ years I've automated workflows that ran for hours down to minutes, shipped computer-vision models north of 97.5% accuracy in production, and built specialized models that outperform frontier LLMs on-task at a tiny fraction of the compute. AUTOMATION I design pipelines and AI-driven workflows that replace manual, error-prone processes: LLM-powered document processing, data ingestion and ETL, RAG/MCP integrations, and agentic systems that handle the busywork. I've cut enough of it for clients to gain multiples worth of efficiency gains. PREDICTIVE ANALYTICS I build forecasting, classification, and risk/scoring models that turn historical data into reliable signal - with the feature engineering, evaluation, and monitoring to keep them accurate in production, not just in a notebook. I also handle the engineering around the models - infrastructure, security, and the supporting services that keep everything running in production. When a project calls for it, I work across NLP, computer vision, and custom model training/fine-tuning. SELECTED RESULTS - Automated document processing from hours to minutes with LLMs - 97.5%+ accuracy on production computer-vision inference - Fine-tuned a compact language model to more than 5x the on-task accuracy of a frontier LLM (GPT-5.5) while using thousands of times less compute CORE SKILLS - Predictive analytics & forecasting, machine learning, deep learning, LLMs, computer vision - Workflow & data automation, MLOps, custom model training, optimization & fine-tuning - Supervised, unsupervised (K-Means, HDBSCAN), and reinforcement learning (Stable-Baselines3) STACK - Languages: Python, SQL, Bash - ML/Data: scikit-learn, PyTorch, TensorFlow, XGBoost/LightGBM/CatBoost, pandas - Build & Deploy: FastAPI/Flask, Docker, Kubernetes, Git I care about modular, efficient systems that keep working long after handoff - not one-off scripts that break the moment requirements shift. If you've got a process worth automating or data worth predicting from, let's talk about moving your project forward.