Kevin L

AI Automation Engineer & Researcher

Previously: Apple, Cisco

Driven by a passion for innovation, I build intelligent systems that push boundaries. With experience shipping products at Apple and Cisco, I've worked across healthcare, finance, startups, and enterprise platforms. Currently focused on AI automation, decentralized systems, and personal infrastructure—exploring how intelligent automation and blockchain can unlock new models for trust, efficiency, and real-world impact. Every challenge is an opportunity to build something meaningful.

GitHub LinkedIn

Current Personal Projects

Chronicle AI

Faster-Whisper Ollama Frigate NVR n8n MQTT OpenCV

Multi-tier AI system for real-time video/audio processing, transcription, and automated life journaling across home infrastructure.

A distributed architecture spanning Raspberry Pi 5 (camera hub), Desktop (AI processing with dual GPUs), and Proxmox server (orchestration). Continuously monitors 3 camera feeds, transcribes audio via Whisper, analyzes video frames, and stores metadata. Uses n8n workflows to generate daily summaries and weekly recaps with Ollama LLM analysis, all queryable through OpenWebUI.

Status: Phase 1 active development (real-time processing implemented)

Architecture Highlights:

  • 7-day rolling video buffer with 1-day retention on Pi
  • Dual-GPU allocation (transcription + vision processing)
  • Event-driven MQTT pipeline for real-time analysis
  • Automated daily batch processing and weekly summarization
  • Modular design for future mobile inputs (RODE mic, iPhone streaming)
  • Fully local, air-gapped AI pipeline — all transcription, vision, and LLM inference runs on-premise with no data sent to external cloud services
  • Sensitive lifelogging data (video, audio, transcripts) never leaves the home network, ensuring complete data sovereignty

Tech Stack: Raspberry Pi 5 (8GB + Hailo-8L AI Hat), Desktop (Ryzen 7 7800X3D, RTX 4070S + RTX 5060Ti), Proxmox, Home Assistant, Frigate NVR, go2rtc, Mosquitto MQTT, OpenCV, Faster-Whisper, Ollama, n8n, PostgreSQL, OpenWebUI

Neo4j AI Memory System (DarkLens AI Assistant)

Neo4j FastAPI Python CrewAI n8n Pydantic

Graph-based memory system for AI assistants that discovers behavioral patterns and builds confidence-weighted knowledge from user interactions.

A sophisticated memory architecture using Neo4j to track user experiences, automatically discover boolean traits (preferences, habits, values), and generate contextually intelligent questions. Features temporal weighting (recent experiences matter more), confidence scoring, and decision tracking with boolean influences. Includes a complete REST API with FastAPI, n8n workflow automation, background task processing, and production-ready error handling.

Status: Private system - In active development

Key Features:

  • Automatic pattern/boolean discovery from natural language
  • Confidence-based state management with evidence tracking
  • Intelligent question generation (sophistication evolves with confidence)
  • Decision logging with causal boolean linkage
  • Production deployment with health monitoring
  • Self-hosted architecture — behavioral profiles, preferences, and interaction history are stored locally with no third-party data sharing
  • Privacy-first design for sensitive personal data; user profiling and pattern analysis remain entirely on private infrastructure

Tech Stack: Neo4j, FastAPI, Python, Pydantic, n8n, systemd, CORS middleware, CrewAI integration

VolunteerChain

Web3 DAO Solana Next.js Rails API

Web3 DAO platform for volunteer organizations built on Solana, enabling decentralized governance and transparent crypto donations.

Full-stack Web3 platform enabling volunteer organizations to form DAOs with decentralized governance and transparent on-chain operations. Built on Solana blockchain with Next.js frontend and Ruby on Rails API backend. Features DAO-based decision making, smart contract integration for crypto donations, project coordination with blockchain verification, and complete transparency through on-chain records. PostgreSQL database with Sidekiq and Redis for background job processing.

Status: In active development

Key Features:

  • DAO formation and decentralized governance for volunteer organizations
  • On-chain crypto donation management with complete transparency
  • Smart contract integration for automated fund distribution
  • Dockerized development environment with hot reloading
  • RESTful API with multi-domain architecture

Tech Stack: Solana blockchain, Web3, Next.js, Ruby on Rails, PostgreSQL, Sidekiq, Redis, Docker

Personal4Cast.ai

AI/ML SaaS FinTech Rails 8

Full-stack SaaS platform for personal finance management with AI-powered insights and comprehensive portfolio tracking.

Production-ready financial management platform built with Rails 8 and modern web technologies. Features OpenAI integration for personalized financial analysis, multi-tenant architecture with complete data isolation, investment portfolio tracking with real-time P&L calculations, expense management, and automated net worth tracking. Includes Stripe integration for subscription billing and comprehensive test coverage with CI/CD pipeline.

Status: In beta - Marketing page live at personal4cast.ai accepting beta signups

Technical Highlights:

  • Multi-tenant SaaS architecture with secure data isolation — strict row-level tenant boundaries ensure users can never access another tenant's financial data
  • OpenAI GPT integration for AI-powered financial analysis
  • Real-time investment portfolio tracking (stocks, crypto, ETFs)
  • Stripe subscription billing with feature access control
  • Comprehensive RSpec test suite with automated CI/CD
  • Production deployment with Kamal and Docker

Tech Stack: Rails 8, PostgreSQL, OpenAI GPT, Stripe, Tailwind CSS, Stimulus.js, Hotwire Turbo, Kamal deployment, Docker, RSpec, GitHub Actions

Marketing Page

PromptMeNot

Ruby Gem Security Rails Open Source

Published Ruby gem for detecting and sanitizing prompt injection attacks in user-submitted text for Rails applications.

Comprehensive prompt injection detection library with ~60 patterns across 6 attack categories: direct instruction override, role manipulation, delimiter injection, encoding obfuscation, indirect injection, and context manipulation. Features configurable sensitivity levels (low/medium/high/paranoid), two operating modes (reject with validation error or sanitize content), and seamless ActiveModel integration.

Status: Published on RubyGems - Production ready

Key Features:

  • ~60 detection patterns across 6 attack categories
  • Filter-based sensitivity levels with configurable activation thresholds
  • ActiveModel validator integration with Rails
  • Standalone API: Promptmenot.safe?, .detect, .sanitize
  • Global configuration DSL with custom pattern support
  • Detection callbacks for logging and monitoring
  • Rails generator for one-command setup
  • I18n support for error messages

Tech Stack: Ruby 3.0+, Rails 6.0+, RSpec, GitHub Actions CI/CD

GitHub RubyGems

KloakOfInvisibility

SwiftUI Swift iOS

iOS app for removing EXIF metadata from images to protect user privacy.

A mobile application that strips location data, camera information, and other metadata from photos before sharing, giving users control over their digital footprint. EXIF metadata can expose GPS coordinates, device identifiers, timestamps, and camera serial numbers — data commonly used for location tracking and device fingerprinting. All metadata stripping is performed entirely on-device with zero network calls, so images never leave the user's phone during processing.

Tech Stack: SwiftUI, Swift, iOS, EXIF metadata manipulation

App Store

Gravemind Studios - AI Game Generator

Godot Engine GDScript Python AI/ML

AI-powered 2D game generation system built with Godot Engine, creating playable games from text descriptions.

Exploring creative AI applications in game development by manipulating Godot Engine programmatically to generate 2D games from natural language prompts. Building machine learning models capable of procedurally generating game assets, mechanics, and level designs, then automatically constructing playable games in Godot. Collaborative project pushing the boundaries of AI-assisted game creation.

Status: In development - Not public

Tech Stack: Godot Engine, GDScript, Python, Machine Learning, Procedural Generation

Gravemind Studios

AI Pet Monitor

Python 3.14 TensorFlow OpenCV

TensorFlow-based computer vision system monitoring dog food and water levels with automated alerts.

Custom-trained object detection model using TensorFlow to track pet bowl status in real-time, sending notifications when food/water needs refilling. Runs entirely on local hardware with on-device inference — no camera feeds or video data are sent to cloud services, keeping home surveillance footage private.

Status: Personal use - Not public

Tech Stack: Python 3.14, TensorFlow, OpenCV, Computer Vision