dhans.AI engineer. Curious by default.Let’s talk
Dean Hans
AI Engineer at egeroo.ai

I build AI that
earns its place.

Production systems. Faster responses. Lower costs.
From the model to the person using it.

PROFESSIONAL EXPERIENCE

Currently building AI at egeroo.ai.

Feb 2026 - PresentFull-time

AI Engineer

egeroo.ai

Improving retrieval, provider integrations and evaluation for production conversational AI.

Delivered7.22s → 1.1s intent-query latency; MLflow tracing across RAG and agentic flows.

Feb 2025 - Jan 2026Internship

AI Engineer Intern

egeroo.ai

Built RAG Q&A, order-taking, recommendation and multimodal agents for conversational products.

DeliveredLowered AI operating cost by about 85% and shipped four chatbot architectures.

The full story

What I delivered at egeroo.ai.

Production improvements across cost, latency, and product iteration.

What I reach for.

Tools follow the problem.

AI systems

PythonLangChainRAGEmbeddingsLLM APIs

Models & data

PyTorchHugging Face Transformersscikit-learnPandasSQL

Analysis & visualization

NumPyMatplotlibSeabornPlotlyStreamlit

Into production

DockerPostgreSQLpgvectorVector searchMLflow
See them at work

Built with a reason.

All projects
Data Science/Analytics

Loan analysis

Analyzed customer data and compared classification models, delivering a notebook, model artifacts and a written report.

Random Forest model accuracy
97.40%
Pythonscikit-learnXGBoost
ML Engineering

Makemore

Built a character-level GPT from scratch using PyTorch, implementing attention, training and Indonesian poetry generation.

parameters built and trained from scratch
10M+
PythonPyTorch
AI Engineering

Dio the Chatbot

A conversational RAG guide that answers questions across Hans’s projects, experience and source material.

to understand Hans quickly
1 chat
LangChainRAGRetrievalEmbeddings

Always a student.

Courses, certificates, and time spent getting better.

All 10 credentials

Research Publication.

RESEARCH · 2025

Can a model learn more
without forgetting?

Extended DynaMMo with class-balanced focal loss to help a model learn new eye-disease classes from imbalanced images.

latest accuracy
58.03% → 69.30%

10 RFMiD classes · 5 incremental tasks · 25-sample memory buffer.

Procedia Computer Science 269 · pp. 649–658

The foundations.

Bachelor of Computer Science
BINUS University

3.90/4.00Magna Cum Laude
Intelligent Systems · 2026

Have something
worth building?

Jakarta / Tangerang, Indonesia

© 2026 Dean HansBuilt thoughtfully. Served with curiosity.
🤗