I build production LLM systems from the metal up — from quantized models running on Jetson edge hardware to multi-agent cloud deployments with tool-use, permission gating, and audit trails. I care about systems that are secure, measurable, and actually useful.

Dallas-Fort Worth, TX

Core stack: Python, Rust, PyTorch, CUDA, Docker, LiteLLM

Security-first open-source coding agent. Hand-rolled async ReAct loop with a 4-tier deny-first permission engine, SHA-256 hash-chained audit trail, and 200+ LLM providers via LiteLLM.

• 30+ built-in tools with JSON Schema validation, MCP server/client
• Parallel + speculative tool dispatch, cost budget enforcement
• Self-evolution via LLM-guided mutations, multi-language verify gate
• SWE-bench Lite: 52.2% oracle best-of-5

Autonomous multi-agent personal intelligence system running on NVIDIA Jetson hardware. Fully on-device inference — zero cloud dependencies, privacy-preserving by design.

Multi-agent algorithmic trading pipeline with DeepSeek R1 reasoning at every stage.

• 4-agent pipeline: Technical Analysis → Chief Strategist → Risk Manager → Execution
• Kelly Criterion position sizing, Monte Carlo risk simulation
• Real-time WebSocket market data, paper trading integration

Qwen3.5-4B fine-tuned with ORPO for biblical question-answering.

• Hybrid RAG: dense embeddings + keyword search
• Constitutional AI self-critique guardrails for theological accuracy
• Voice pipeline: speech-to-text → LLM → text-to-speech

Comprehensive GPU diagnostic toolkit modeled on NVIDIA DCGM architecture.

• Automated stress testing, memory validation, ECC detection
• Health monitoring for GPU server fleets

Production-grade ML training infrastructure for single-GPU homelabs.

• Unsloth fp8 quantization, torch.compile graph optimization
• DeepSpeed ZeRO stages, vLLM + lm-eval harness
• Multi-seed reporting for statistically sound results

SQL + Python ETL pipeline for semiconductor quality analysis.

• Supplier performance scoring with trend detection
• Defect Pareto distributions, yield rate dashboards
• Automated alerting on quality threshold breaches

Multi-model ML pipeline predicting tire wear for Tesla vehicles. Random Forest, XGBoost, Neural Network, and Ensemble models with Claude AI analysis for tire longevity insights.

• Simulated driving data with vehicle-specific tire degradation modeling
• GridSearch-tuned Random Forest, XGBoost, and TensorFlow/Keras neural network
• Ensemble averaging across all models for robust predictions
• Claude AI integration for natural language tire wear analysis

Git-backed knowledge wiki — Karpathy's LLM Wiki pattern with LangGraph ingestion pipelines for structured and unstructured content. Full diff history.

📈 Contribution Graph