Featured Projects

Here are some of the projects I've been working on recently.

Fayetteville PWC

AI and Data Engineer at Fayetteville Public Works Commission: a centralized data lake and pipelines that put operational data in one queryable place, BI reporting on top of it, custom RAG agents that let staff search internal documentation and the data dictionary, purpose-built agents in Microsoft Copilot Studio and Oracle AI Agent Studio for business teams, and hands-on AI training for staff across the utility.

PythonSQLOracle DatabaseRAGMulti-Agent RAGEmbeddingsVector DatabasesMicrosoft Copilot StudioOracle AI Agent StudioPrompt EngineeringAI TrainingStaff TrainingAPI DevelopmentAPI Integration

RFP Pilot

Co-founder of RFP Pilot, an AI proposal tool for state and local government and disaster-recovery responders. It reads the raw solicitation as posted, extracts every requirement into a traceable compliance checklist, and drafts first responses grounded in the firm’s own past proposals, citing sources, scoring confidence, and flagging gaps instead of fabricating. I built the document ingestion and semantic retrieval pipeline, the RAG grounding, and the human-in-the-loop review flow that lets reviewers validate question by question, lowest-confidence answers first. Running free concierge pilots with early customers.

rfppilot.com
PythonTypeScriptNext.jsSQLMulti-Agent RAGOpenAI APIVector DatabasesPostgreSQLpgvectorAWSLambda FunctionsAPI DevelopmentAPI IntegrationGitFrontend Design

pnwoods.com + private AI workspace

This site, plus an invite-only ChatGPT-style workspace at ai.pnwoods.com that runs open-weight coding models on my own hardware. Next.js on Vercel, Supabase for auth, Postgres and file storage, hybrid vector + full-text RAG over uploaded documents and a full data dictionary, per-user memory and personalization, live web search through a self-hosted SearXNG, and Ollama exposed through a Cloudflare Tunnel locked with Zero Trust service tokens. The same model powers an agentic coding setup in VS Code.

Source on GitHubai.pnwoods.com (invite only)
TypeScriptNext.jsReactTailwind CSSVercelSupabasePostgreSQLpgvectorHybrid SearchEmbeddingsRAGOllamaLocal LLM HostingPrompt EngineeringModel Context Protocol (MCP)Cloudflare Tunnel & Zero TrustDockerProxmoxTailscaleSelf-Hosted Services

Home lab: Proxmox + self-hosted services

A Proxmox VE host running the services behind everything else here, with local LLM inference on 96 GB of pooled GPU memory across three NVIDIA cards and an Apple silicon machine. Each service lives in its own LXC container: Pi-hole DNS for the whole network, the SearXNG metasearch engine that gives the AI workspace live web results, a Docker media-automation stack, a network file share, two dedicated game servers, and a small game-community bot. Tailscale handles remote access, and the few public endpoints go out through Cloudflare Tunnel behind Zero Trust.

ProxmoxLinux/UnixDockerDocker ComposeBashTailscaleCloudflare Tunnel & Zero TrustOllamaSelf-Hosted ServicesNetwork AdministrationNAS ManagementSSH

Clemson AI Research - Edge LLMs

Undergraduate AI research at Clemson University on optimizing and deploying Large Language Models on edge devices. Co-authored “Don't Waste Bits! Adaptive KV-Cache Quantization for Lightweight On-Device LLMs”, accepted to the LoViF Workshop at CVPR 2026. Contributed the base test suite and built the learned controller with Gabriel Hillesheim: it selects between 2-bit, 4-bit, 8-bit and FP16 precision per token using lightweight signals to improve the accuracy-latency trade-off.

Read the paper on arXiv
PythonPyTorchHugging FaceLangChainLLM OptimizationEdge ComputingModel CompressionQuantizationPruningKnowledge DistillationParameter-Efficient Fine-Tuning (PEFT)Inference OptimizationTechnical Presentations

Clemson AI Research - RAG Systems

Undergraduate AI research at Clemson University using the Palmetto HPC cluster for AI training and optimization. Developed a RAG model combining retrieval-based search with generative AI to improve response accuracy, and presented the work to University of Florida faculty.

PythonLangChainHugging FaceOpenAI APICUDA ToolkitRAGEmbeddingsParallel ComputingGPU ProgrammingDistributed ComputingSlurmPalmetto Cluster HPCAI TrainingTechnical Presentations

BCDA LLC Network Infrastructure

Led network expansion and software implementation at BCDA LLC: grew the network to support 42 new devices, deployed and trained all staff on new software systems, and kept 99.9% network uptime with 95% of technical issues resolved within 6 hours.

Network AdministrationSystem IntegrationInfrastructureNAS ManagementSSHWindows ServerVMware VirtualizationDocker ComposeStaff TrainingTeam LeadershipTechnical Support

Want the details on any of these? Email me.