Skip to content

~/developer/julius

Julius Narvasa

Software Developer

I build backend systems, self-hosted AI infrastructure, and developer tooling. Currently exploring local LLMs and computer vision pipelines on AMD hardware.

backend · apis · cloud infra · local ai · self-hosted systems

01 — Projects

Featured Projects

Things I've built end-to-end.

Re-Fuel mobile app home screen showing gas station price tracking

Re-Fuel

mobile

Drivers manually log gas prices and fuel consumption across the Philippines — no easy way to see trends or remember which station was cheapest last week.

Challenges

  • GPS-based auto-capture within 150m
  • On-device YOLOv8 price-board detection (ONNX)
  • Offline-first IndexedDB cache
react capacitor typescript yolov8 onnx csharp dotnet-8 postgres
AI Resume Builder upload step showing file drag-and-drop interface

AI Resume Builder

web

Tailoring a resume to every job posting is tedious and error-prone. Generic resumes get filtered out.

Challenges

  • PDF + DOCX resume parsing
  • Playwright-based LinkedIn scraping
  • Structured LLM output for tailored CV + cover letter
fastapi python react vite tailwindcss playwright openai-api pymupdf
GpuVramMonitor showing per-process VRAM usage on AMD Radeon RX 7800 XT

GpuVramMonitor

desktop

Windows has no per-process VRAM tool for AMD GPUs. Task Manager shows only the total; AMD Adrenalin shows only the total.

Challenges

  • Parsing GPU Process Memory performance counter instance names
  • Resolving PIDs and excluding the inflated dwm.exe reading
  • Real-time refresh with pause/resume
csharp wpf dotnet-8 windows performance-counters

02 — Infrastructure

Infrastructure

Self-hosted across a Proxmox box at home and a cloud VPS.

Homelab

Proxmox · on-prem
├─ VM 100 · apps
├─ Immich (Docker)
├─ immich_server
├─ immich_postgres
├─ immich_redis
└─ immich_ml
├─ Jellyfin (native)
├─ qBittorrent (native)
└─ Ollama (native)
├─ VM 101 · networkwatcher
├─ VM 102 · OmadaController
└─ VM 103 · autoswitcher

VPS (cloud)

VPS · cloud (ARM64)
├─ Nginx reverse proxy (public-facing)
├─ SillyTavern (node)
├─ AI Resume Builder (FastAPI)
├─ Re-Fuel API (.NET)
├─ Ollama — embeddings
├─ vaultwarden
├─ n8n
├─ PostgreSQL
├─ Redis
└─ WireGuard

Local workstation

Local workstation
├─ llama.cpp + Qwen3-Coder (LLM inference)
└─ ComfyUI (image gen)

Networking

Networking
├─ Tailscale (mesh)
├─ WireGuard
├─ Cloudflare DNS
└─ Nginx reverse proxy

Cloud

Cloud
└─ Oracle Cloud (ARM64)

03 — AI Lab

AI Lab

Local inference and computer vision experiments.

Online — local inference ready

Inference

llama.cpp b9106
Qwen3.6-35B-A3B Q4_K_M
AMD ROCm / HIP gfx1101

Vision

ComfyUI ROCm
Chroma1-HD Q6_K

04 — Experience

Experience

Where I've been and what I've shipped.

2020

Joined Netzon Tech

Backend dev, internal tooling, observability.

2021

Led API development

Multi-tenant ISP management platform (GraphR).

  • Stripe + GoCardless integration
  • Subscription recovery workflows
2022

Cloud integrations

Azure + GCP cost optimization, IaC, CI/CD.

  • Cut cloud spend 30% via right-sizing
  • Migrated monolith to managed services
2023

Payments & billing

Led billing architecture across products.

  • Idempotent webhook handling
  • PCI-conscious design
2024

AI & computer vision

Started Re-Fuel price-board detection.

  • YOLOv8 on-device via ONNX
  • OCR preprocessing pipelines
2025

Local AI infrastructure

Self-hosted inference + image gen.

  • llama.cpp + ROCm on RX 7800 XT
  • ComfyUI for image generation
2026

Exploring

Local assistant ecosystem + payment automation.

  • Multi-backend proxy + embedding server
  • n8n + Stripe workflows

05 — Now

Currently Building

Active threads. Updated as things ship.

  • Local AI assistant ecosystem

    multi-backend proxy + embedding server

    wip
  • Fuel-price recognition platform

    end-to-end, beyond the Re-Fuel MVP

    research
  • Self-hosted AI infrastructure

    HIP-native builds, draft-model speculation

    shipping

06 — Contact

Get in touch

Open to interesting backend, infra, and AI work.