Projects
Innovative solutions leveraging AI, enterprise architecture, and modern technologies to solve real-world challenges.
NutritionAI
AI-powered nutrition tracking and personalized meal planning application that helps users achieve their health goals through intelligent food analysis
Key Features
- AI-powered meal tracking via image analysis
- Personalized nutrition recommendations based on health goals
- Detailed nutritional breakdowns (macros and micros)
- Daily progress tracking and insights dashboard
- Recipe suggestions and meal planning
SAP Solution Advisor
Web-based, mobile-friendly application designed to help SAP consultants and partners identify the best SAP solutions for specific business challenges
Key Features
- Search, filter, and compare various SAP solutions
- PDF matching to find similar solutions based on uploaded documents
- Manage favorites and add personal notes
- Rate solutions for future reference
- Mobile-friendly responsive design
Mac Mini + ASUS Security & Kids Usage Node
Production blueprint and software suite transforming an always-on Apple Silicon Mac Mini into an autonomous, local-first security appliance, CrowdSec firewall bouncer, and privacy-preserving kids web usage monitoring node for ASUS AiMesh routers with TypeSafe Jev AI integration
Key Features
- TypeSafe Jev AI integration for automated domain safety classification, real-time threat categorization, and natural language intent routing with circuit-breaker fallback
- Local-first kids browsing sessionization engine calculating true active screen time vs. idle hours without exporting telemetry to the cloud
- Hardware MAC address anonymization matching devices locally to human-friendly aliases (e.g., Kids Laptop)
- Automated network intrusion prevention syncing CrowdSec threat decisions into ASUS router iptables/ip6tables DROP chains via SSH
- Multi-tier resilient SQLite architecture with WAL index auto-recovery, memory probes, and immutable zero-lock fallbacks
- Unified telemetry access via 3-layer authorized Telegram bot (/kids, /status, /check) and custom Stdio MCP server for AI assistants
LocalSafeKidsEngine
Open-source hybrid kids traffic analysis and safety classification PoC combining local Apple Silicon M4 Small Language Models (SLMs) with TypeSafe.AI (Jev) frontier System One decision calibration
Key Features
- Multi-tier hybrid architecture routing traffic between microsecond fast-filters, local on-device SLMs, and cloud calibration
- Apple Silicon M4 local inference running 4-bit quantized SLMs in <1.85 GB unified RAM at 10-15 ms latency on Metal GPU
- TypeSafe.AI (Jev) System One integration for non-autoregressive decision calibration on borderline queries (<0.88 confidence)
- Age-based multi-tier policy engine supporting customizable profiles for Tots (3-6), Kids (7-11), and Teens (12-16)
- On-device LoRA fine-tuning workflows via Apple MLX and exportable 4-bit GGUF quantization using Unsloth