💻 Tech & Projects

Retrieval-augmented AI assistant that helps real estate brokers query documents and get instant, context-aware answers using a locally-hosted LLM.

• Built a RAG pipeline connecting a local Ollama model to a custom document knowledge base
• Express backend + React frontend, deployed publicly via Cloudflare Tunnel on a custom subdomain
• Created zsh aliases and systemd services for one-command deployment and management
• All inference runs locally — no third-party API costs or data exposure

Real-time multiplayer web app that teaches optimal blackjack strategy through live play and instant feedback.

• Socket.io-based real-time multiplayer engine supporting multiple concurrent tables
• Basic strategy logic evaluates player decisions and gives instant correctness feedback
• React frontend for live game state, hand visualization, and strategy hints
• Self-hosted with custom routing, sessions, and game state persistence

Automated two-way SMS system connecting a locally-hosted LLM to Google Voice via browser automation — no official API needed.

• Built undetected-chromedriver automation to handle Google Voice send/receive
• Integrated Ollama LLM backend with persistent multi-turn conversation history
• Custom persona/character card support for swappable AI identities at runtime
• Configurable response filtering and safety layers

Interactive web app visualizing Seattle-area housing market trends by neighborhood, built on King County public property data.

• FastAPI backend serving median price trends, days-on-market, and inventory data
• React + Plotly frontend with interactive neighborhood filtering
• Rental ROI calculator with cap rate and cash-on-cash return metrics
• Market temperature scoring system ranking neighborhoods hot/warm/cool/cold

Full personal computing stack built and maintained on Arch Linux — from kernel drivers to desktop automation across a triple-monitor setup.

• Custom Arch Linux setup with NVIDIA driver config and i3wm multi-monitor layout
• Local LLM inference stack using Ollama on an RTX 3060 (llama3.1, qwen2.5-coder)
• Automated system tasks via Bash and Python: theming pipelines, service management
• Resolved NVIDIA/Wayland conflicts migrating from Hyprland to X11/i3

Self-hosted Paper Minecraft server for a community of kids — designed so children could play together regardless of device, with full crossplay between Java and Bedrock editions.

• Enabled crossplay via BedrockConnect so kids on Switch, mobile, and PC could all join the same server
• Configured LuckPerms RBAC permission system with child-safe role hierarchies
• Integrated PlaceholderAPI with leaderboard plugins for live player stats
• Python scripts using amulet-nbt to enable experimental features via NBT manipulation
• Managed the server remotely via SSH, keeping it stable and online for the community

Local Stable Diffusion pipeline for generating tattoo stencils and previewing how a design would actually look on skin before committing to it.

• Custom ComfyUI workflows for iterative, style-consistent tattoo stencil generation
• Compositing pipeline to preview generated designs placed on a body part for realistic visualization
• Optimized inference parameters (CFG scale, samplers, LoRA weights) for clean, targeted line work