AI Engineer in Semarang, Indonesia. I build AI systems from the infrastructure and service architecture side: model gateways, request routing, failure handling, and tool orchestration.
I am studying pure mathematics at Universitas Negeri Semarang. The habit of working from definitions and proofs carried over, so I read a model's assumptions and limits before putting it into service.
Website · LinkedIn · mrpandanarang@gmail.com
- Model infrastructure — gateways, provider routing, quota and cost control
- Applied artificial intellegence — evaluation design, leakage checks, reporting limits honestly
- Backend services — FastAPI, Docker, Caddy, SQLite, GitHub Actions
| Project | What it does | Stack |
|---|---|---|
| OmniGate | One OpenAI-compatible endpoint that pools several free-tier LLM providers, with fallback and routing driven by measured quality, latency, and remaining quota. | Bun, TypeScript, Docker |
| NeuralAlpha | Two-stage portfolio optimisation for LQ45 Indonesian equities: masked-autoencoder pre-training, CNN-BiLSTM walk-forward preselection, then mean-variance allocation under real IDX retail costs. | Python, PyTorch |
| OpenData-Jatim | East Java economic intelligence: Composite Risk Score modelling with interpretable XGBoost surrogates and LLM policy narratives. Finalist, Jatim Datathon 2025. | Python, XGBoost |
| kalshi-diffusion-tracker | Measures how quickly news is absorbed into Kalshi prediction-market prices, with Brier-score calibration checks. | Python |
Founder of Phi Consulting, an independent AI engineering practice that takes machine learning and automation projects end to end for Indonesian companies without an in-house AI team.