Real builds. Zero cloud calls. Working software speaks.
Print & Graphics · File Intake
Print & Graphics File Intake Agent
FastAPI + local LLM pipeline for automated file intake, classification, and routing. Zero external API calls. Built and demoed live in under a week.
// SQLite · PyMuPDF · Ollama · stdlib only
Finance · Agentic Trading
Agentic Trading System
Local LLM-driven trading agent with human-approval flow and live broker API integration. Demoed live at the Anaheim Chamber of Commerce mixer.
// FastAPI · Qwen3 · Human-in-the-loop approval
Legal · Document Intake
Legal Document Intake Agent
Sovereign document intake pipeline for law firm clients. Classifies, extracts, and routes locally. Per-client isolation, zero data sharing between instances.
// Air-gapped · per-client isolation · HIPAA-ready
Hospitality · AI Chat
Hospitality Guest AI System
Fully sovereign AI chat system for a short-term rental platform. FastAPI + Ollama + SQLite backend, Notion as admin input layer, emergency keyword filter, and mobile-first guest UI. Zero external API dependencies.
// FastAPI · llama3.2:3b · zero cloud dependencies
Aerospace · Facility Intelligence
Facility Mapping & Floor Plan Pipeline
Air-gapped pipeline converting LiDAR point clouds to DXF floor plans for ITAR-sensitive facilities that cannot use cloud SaaS. Hand-rolled DXF writer, numpy-based processing, LLM-powered ADA compliance narration.
// numpy · DXF · Qwen2.5:32b · ITAR-safe
Edge AI · Flight Controller · R&D
Arduino VENTUNO Q Edge AI Platform
Next-generation drone brain — Arduino VENTUNO Q with 40 TOPS NPU, integrated STM32 MCU, and 16GB RAM replacing the current Pi Zero 2W + Matek FC stack. Purpose-built for fully autonomous, on-device AI inference at the edge with zero cloud dependency.
// 40 TOPS NPU · STM32 · 16GB RAM · air-gapped autonomous flight
Edge AI · Heavy Inference · R&D
Jetson Orin Nano Super — Edge Inference Node
NVIDIA Jetson Orin Nano Super Developer Kit deployed as a dedicated heavy inference node in the MotherBrain cluster. Sovereign on-device AI for autonomous platforms — no cloud, no egress, ITAR-safe edge compute for aerospace and defense applications.
// Jetson Orin Nano Super · on-device inference · MotherBrain cluster
Medical · Protocol Assistant
Medical Device Protocol Assistant
On-premise AI assistant covering 13 machines and 35+ treatment protocols for a medical device client. Full four-file sovereign stack: FastAPI + Ollama + SQLite + vanilla JS. Live-demoed via secure tunnel.
// 13 machines · 35+ protocols · air-gapped stack
Voice AI · Conversational Agent
Sovereign Voice Assistant
Fully local voice AI pipeline — browser push-to-talk → Whisper STT (GPU-accelerated) → local LLM → Piper neural TTS → spoken response. No cloud, no API keys, no per-minute billing. Full conversation history, HTTP auth on admin routes. Direct alternative to commercial voice AI subscriptions that send data offsite.
// FastAPI · Whisper · Piper TTS · llama3.2:3b · CUDA · SQLite
Infrastructure · Token Optimization
Headless Token Reduction Pipeline
Headless browser pipeline that pre-processes and strips web content before LLM ingestion, dramatically reducing token count and inference cost. Extracts only semantic signal — no markup, no noise — enabling leaner, faster, and cheaper local model inference at scale.
// Headless · token reduction · local inference optimization
Edge AI · Robotics · Hackathon
Autonomous Soccer Robot
Fully autonomous soccer robot built at the AI LA/Qualcomm Robot Soccer Cup hackathon on an Arduino UNO Q (QRB2210) with Edge Impulse FOMO object detection. Working navigation and ball-tracking in a single session — zero cloud, pure edge inference.
// Arduino UNO Q · Edge Impulse FOMO · on-device inference
Edge AI · Companion Bot · R&D
2B — AI Companion Bot
Sovereign voice AI companion running on a Raspberry Pi Zero 2W edge client streaming to MotherBrain for STT/LLM/TTS. Wake word detection, animated OLED eyes with mouth sync, audio-reactive LED ring, hallucination suppression, and stateless persona — fully local, zero cloud.
// Pi Zero 2W · Whisper · Piper TTS · openWakeWord · SSD1306 OLED
Edge AI · Multi-Agent · Infrastructure
A2A Agent Orchestration Layer
Stdlib-only Python Agent-to-Agent (A2A) prototype on MotherBrain — zero pip dependencies, thread-safe concurrent agents on parallel ports. Research and coding agents running simultaneously, architected to extend across the full child-rig cluster.
// stdlib only · ThreadingHTTPServer · zero dependencies · race-condition safe
Edge AI · Autonomous Systems · R&D
Autonomous AI Drone Platform
Sovereign AI drone built on a split-architecture: local FastAPI inference server (MotherBrain) paired with an onboard Pi Zero 2W flight controller. MAVLink bridge, asyncio state machine, and OpenMV H7 thermal camera — fully autonomous, zero cloud, ITAR-safe by design.
// Pi Zero 2W · MAVLink · thermal vision · air-gapped
