Johannes Baptista Adiatmaja Pambudi

|

AI Native Engineer building intelligent systems through agentic AI, RAG, and deep learning.

Scroll

About

AI Native Engineer based in Indonesia, building end-to-end AI solutions from agent orchestrators to production models.

Education B.S. Informatics, UKDW GPA 3.75/4.00 — 2024
Certification
TensorFlow Developer Google — 2023

Skills

Python
SQL
MySQL
Java
TypeScript
PHP
PyTorch
TensorFlow
Keras
scikit-learn
Pandas
NumPy
Streamlit
ChromaDB
Docker
Git
n8n n8n
Python
SQL
MySQL
Java
TypeScript
PHP
PyTorch
TensorFlow
Keras
scikit-learn
Pandas
NumPy
Streamlit
ChromaDB
Docker
Git
n8n n8n

Experience

AI Engineer

Sigma Solusi Indonesia

Aug 2024 — Present
  • Built internal AI assistants (WhatsApp and web-based) with custom data pipelines that let non-technical teams query sales, inventory, and procurement data in plain language across multiple darkstores, eliminating manual SQL reporting.
  • Developed a production demand-forecasting system covering 600+ products that predicts daily sales and revenue and retrains itself automatically each week. Cut chronic under-forecasting by more than half through A/B-tested bias correction with no loss in accuracy, reducing stockouts during peak-demand periods. Built on tiered NHITS and LightGBM models with probabilistic output.
  • Designed exploratory LLM systems including Retrieval-Augmented Generation (RAG) pipelines with semantic retrieval, and fine-tuned domain-specific language models using synthetic instruction generation and knowledge distillation.
Read more →

Machine Learning Engineer

Beehive Drones Intern

Aug 2022 — Jan 2023
  • Developed a drone image-processing pipeline for carbon estimation, improving performance by 93% over the previous baseline.
Read more →

Projects

Professional Work

Systems built in my data science role. The source is private — these cards explain what they do.

2024–2026

FMCG Demand Forecast

Production Demand Forecasting

Forecasts 600+ products across 4,000+ product-warehouse series 30 days ahead on ~8 years of live data, using tiered NHITS + LightGBM models with P10/P90 bands and unattended weekly GPU retraining.

NHITS Probabilistic Forecasting LightGBM MLOps
Read more →
2026

LLM Distillation Pipeline

Indonesian-Javanese Language Model Fine-tuning

Distilled a 70B teacher into a 9B student model through a three-zone Docker pipeline running on a restricted GPU server with no Python runtime.

Knowledge Distillation LoRA Unsloth Docker
Read more →
2025

ChromaDB RAG Chatbot

B2B FMCG Semantic Search & Intent Classification

A Dockerized RAG system on ChromaDB and sentence-transformers — semantic product search over colloquial Indonesian names plus 35+ intent types.

RAG ChromaDB sentence-transformers Docker
Read more →

Personal & Academic

Open-source side projects and research, unrelated to my day job.

2026

Praxis

Govern Any Coding Harness from Inside the One You Already Use

Orchestrator for the execution phase of spec-driven development: wired into your assistant over MCP, it decomposes plans to fit the worker model that implements them, dispatches to pluggable harnesses (OpenCode, Antigravity) in disposable containers, and gates every change through verify, review, and a human merge gate. Never a blind dispatch.

Agent Orchestration Spec-Driven Development MCP Docker
Read more →
2026

Crypto Swing Trading Bot

Fully Deterministic BTC/ETH Trading on Binance

An LLM-assisted swing bot whose decisions are 100% deterministic TypeScript — the LLM only routes commands. Backtested across 10 market regimes: profits in clean trends (+6.52%) while keeping max drawdown under ~5% through every crash.

Algorithmic Trading TypeScript Risk Management Backtesting
Read more →
2024

Gender Classification

Fingerprint Analysis — Undergraduate Thesis

Compared three Gabor filter configurations against raw fingerprints using a custom CNN and 5-fold cross-validation on two datasets, achieving 7% F1-score improvement with the default parameter set.

CNN Gabor Filters TensorFlow K-Fold CV
Read more →
2023

ASAH

Waste Classification Mobile App

Achieved 94% accuracy using MobileNetV2, optimized for real-time inference on Android.

MobileNetV2 TensorFlow Android
Read more →

Contact

Open to opportunities and collaborations.

adi@pambudi.com