CFO-RetinaNet: Convolutional Feature Optimization
archivedConvolutional Feature Optimization for Oil Palm Ripeness Assessment in Precision Agriculture, utilizing RetinaNet architectures for computer vision-based quality grading.
A showcase of things I've built and maintained.
Convolutional Feature Optimization for Oil Palm Ripeness Assessment in Precision Agriculture, utilizing RetinaNet architectures for computer vision-based quality grading.
Developed an end-to-end computer vision pipeline using fine-tuned CNN architectures (ResNet50 and EfficientNet) for automatic detection and classification of brain tumors from MRI brain scans.
Production-grade self-hosted homelab on Proxmox VE 8.x with VLAN-segmented MikroTik networking. Runs 15+ Docker Compose stacks (Jellyfin, Immich, Authentik SSO, Homepage, ArrSuite) on a dedicated Ubuntu VM managed by Portainer EE. HAProxy with Let's Encrypt DNS-01 terminates SSL for all services. Cloudflare Tunnel provides zero-port-forwarding public ingress. A Ugreen NAS DH2300 (Debian 12) provides NFS storage. Tailscale VPN enables secure remote administration.
This portfolio — built with Astro v6 and React 19. Features a glassmorphism dark design system, interactive canvas particle hero, SVG network topology with dynamic tooltips, Shiki dual-theme syntax highlighting, real-time blog search, and a full Playwright E2E test suite (66 tests across 5 browser targets). Deployed via a Docker + Nginx container on a self-hosted Proxmox homelab, with automated GitHub-driven image rebuilds.
Centralized homelab automation pipeline using PowerShell and Bash. Automates VM/container backups via Proxmox vzdump, scheduled system updates across all nodes, and Docker Compose stack deployments via SCP + parallel SSH. Custom Telegram bot integration pushes real-time alerts for service health, failed jobs, and uptime anomalies detected by Uptime Kuma. Portainer and Jellyfin stacks are managed through this pipeline.