Software Developer

Dominikus
Andika

Building at the intersection of software, AI / ML, IoT, and business. I write code, think about systems, and occasionally write about what I learn.

AI / MLIoTWeb DevelopmentBusiness
01 / About

I am a software developer based in Indonesia with a particular interest in the intersection of artificial intelligence, embedded systems, and the business logic that determines whether either of those things matter.

I work across the stack, from training ML models and wiring up IoT sensor networks to building the web interfaces that make them usable. I care about systems that work reliably, code that communicates clearly, and products that solve real problems rather than interesting ones.

Outside of building things, I read a lot, mostly about technology, epistemology, and how businesses actually work versus how they are described. I write about what I find useful or surprising on this blog.

Languages
PythonTypeScriptJavaScriptGoJava
AI / ML
PyTorchHuggingFaceFastAPILangChainYOLOPaddleOCRRAGlocal-llmQwen
Web
Next.jsReactNode.jsTailwind CSSPostgreSQL
IoT
ArduinoESP32
AI Agent Stack
CodexClaudeAntigravity
Tools
GitDockerLinuxVercelCloudflare
02 / Selected ProjectsView all
  • A local RAG system for querying personal notes and documents in Bahasa Indonesia. Runs fully on-device with Ollama, qwen2.5:3b, and nomic-embed-text. No cloud, no API costs, no English-only constraint.

    PythonOllamaqwen2.5:3bnomic-embed-text
  • A local-first CLI note app built in Rust. One binary runs on laptop and Android via Termux, no account, no sync server, no monthly fee.

    RustCLITermuxAndroid
  • A document intelligence pipeline that extracts structured JSON attributes from PDFs, combining Tesseract OCR, semantic retrieval, and GPT-4o to keep extraction accurate without burning API tokens.

    PythonTesseractGPT-4oOpenAI API
  • A full-stack project management platform built for a service company, shelved midway. What I learned about shipping, teams, and POC-first thinking changed how I work permanently.

    Next.jsPostgreSQLTailwind CSSshadcn/ui
  • A proof-of-concept object detection model for company asset recognition, built when the client could only provide 3 training photos. Used Gemini AI to expand to nearly 100 realistic training samples, Roboflow for labeling, and Google Colab's free GPU for training.

    PythonYOLOv8RoboflowGoogle Colab
  • Sistem pencatatan order yang berjalan di atas perangkat yang sudah dimiliki, laptop, wifi, dan HP. Dibangun untuk membantu pemilik bisnis menengah mengontrol pesanan, stok, dan pemasukan tanpa biaya cloud atau langganan bulanan.

    PythonFastAPISQLiteHTML
03 / WritingView all
© 2026 Dominikus Andika