Who I am

Evint
Leovonzko

Open to opportunities — Vancouver, BC

UBC researcher and co-founder building at the intersection of machine learning, computational biology, and open-source infrastructure. Originally from Indonesia; currently making LLMs less painful to run locally and teaching machines to read RNA.

13+
Research papers & analyses
2
Companies founded / co-founded
1K+
Students trained (cloud / AWS)
OSN
Indonesia National Science Olympiad finalist

The story so far

I grew up in Indonesia competing in national science olympiads — the kind of thing that teaches you to be comfortable with hard problems and uncomfortable with easy answers. That instinct helped me earn a place in the inaugural cohort of Beasiswa Indonesia Maju (now Beasiswa Garuda), a full government scholarship by Indonesia's Ministry of Education for outstanding students to study abroad. That brought me to Vancouver, where I enrolled at UBC and immediately started looking for problems worth solving at the intersection of data, biology, and systems.

My first published work applied deep learning and Gaussian blur preprocessing to circular RNA classification, achieving 75% accuracy with 0.14ms inference — fast enough to be clinically useful. The paper is published in URNCST Journal. Around the same time I won Best Research Project at UBC Vantage College's Capstone Conference for work on P-53 tumor suppressor gene mutation analysis using genetic algorithms. Two papers, two directions: one molecular, one computational.

“I chase problems where data hides something meaningful.”

On the infrastructure side, I co-founded Kolosal AI — an open-source platform for running large language models locally, privately, and fast. The desktop app and CLI together have earned over 900 GitHub stars. Before that, I co-founded Genta Technology — the experience that first planted the idea of making AI infrastructure genuinely accessible. After the olympiad years, I also took on an AWS / Cloud Computing trainer role, delivering courses to 1,000+ students across Indonesia.

My research portfolio has since expanded into ML clustering (SOM-TSK, GRASP), gene regulatory network inference, and most recently computational economics — building agent-based Keynesian models in Rust to study Minsky dynamics, zero lower bound persistence, and coordination failure. When I'm not debugging loss curves, I'm reading about macroeconomics or hiking the trails around Vancouver.

Work & ventures

2023 Present
Co-Founder
Kolosal AI — Vancouver, BC

Building open-source infrastructure to make running LLMs locally simple, private, and fast. The Kolosal desktop app and CLI have together earned 900+ GitHub stars. Products span a C++ inference server, AutoML platform with Optuna + MLflow, a model memory calculator, and a Retrieval Management System.

C++ Rust TypeScript PyTorch CUDA Docker FastAPI
2023 Present
Undergraduate Researcher

First-author research in computational biology (circular RNA, GRN inference), ML systems (SOM-TSK, GRASP), and computational economics (Keynesian ABM). Published in URNCST Journal; Best Research Project at UBC Vantage College Capstone Conference 2023.

Deep Learning Genomics Agent-Based Modelling Rust Python
2022 2023
Co-Founder
Genta Technology — Indonesia

The early seed of what would become Kolosal AI. Built the Genta Academic Assistant — an AI-powered paper search tool using Streamlit, Weaviate, and the OpenAI API — which first crystallised the idea of making AI infrastructure accessible and locally deployable. Also served as AWS / Cloud Computing trainer, delivering courses to 1,000+ students across Indonesia.

Python Streamlit Weaviate OpenAI API AWS

Education

2023 — Present
University of British Columbia
BSc Computer Science — Vancouver, BC

Enrolled via UBC Vantage College. Active undergraduate researcher in deep learning and genomics. First-author URNCST publication (2024). Best Research Project, Capstone Conference 2023.

Pre-2023
Indonesia — High School
National Competitions & Research

National Science Olympiad (OSN/KSN) finalist in Informatics. Also competed in the national debate circuit — LDBI (Lomba Debat Bahasa Indonesia) and NDDC (National Schools Debating Championship). Conducted independent research projects in high school alongside these competitions.

Cohort 1
Beasiswa Indonesia Maju
Indonesian Government Full Scholarship — Now: Beasiswa Garuda

Full government scholarship awarded by Indonesia's Ministry of Education, Culture, Research & Technology to outstanding students for undergraduate study at international universities. Among the inaugural cohort of recipients.

Awards & honours

Beasiswa Indonesia Maju (Beasiswa Garuda) — First Cohort
Indonesian Government Full Scholarship · Ministry of Education, Culture, Research & Technology · Awarded to high-achieving Indonesian students for undergraduate study abroad · Cohort 1 recipient
Best Research Project — UBC Vantage College Capstone Conference
2023 · P-53 Tumor Suppressor Gene Mutation Analysis using Genetic Algorithms · University of British Columbia
National Science Olympiad Finalist — Informatics
OSN / KSN · Indonesia · Competed against top students nationwide in computational thinking and programming
First-Author Publication — URNCST Journal
2024 · “Deep Learning with Gaussian Blur Preprocessing for Circular RNA Classification” · Vol. 8 · DOI: 10.26685/urncst.601

Selected publications & papers

01
Deep Learning with Gaussian Blur Preprocessing for Circular RNA Classification and Detection
URNCST Journal, Vol. 8 — July 2024
k-mers feature extraction + ANN classifier. Accuracy 75.1%, precision 79.8%, inference 0.14 ms/prediction. Co-authors: Cahyaningrum, Ulwani (UBC Chemistry).
Published
02
Fiscal Stabilisers, Minsky Dynamics, and Distributional Outcomes in a Keynesian Agent-Based Model
Target: Journal of Economic Dynamics and Control
10,000-agent daily-frequency ABM in Rust. ZLB binds in 93.7% of MC periods — a novel result mirroring post-1998 Japan and post-2013 Euro-area.
Preprint
03
Individual Optimality and Collective Failure in a Keynesian Agent-Based Model
Target: Journal of Economic Dynamics and Control
Grid search identifies survival-maximising strategies. All-best economy: GDP −65%, firm bankruptcies 17× higher, mean profit negative — a structural coordination failure attractor.
Preprint
04
GRASP: Graph-Routed Adaptive Spectral Partitioning
First-author — Clustering / ML
Parameter-free clustering pipeline. GNG → spectral gap → oracle routing to 6 specialists. ARI 0.378 on interlocking rings; 50K points in 140ms.
Preprint
05
SOM-TSK: Topology-Seeded K-Means Clustering Framework
First-author — Clustering / ML
Three SOM-based seeding strategies + DenSOM + AutoSOM k-selection. 6 wins, 0 losses vs KMeans++ across 24 benchmark datasets.
Preprint
06
Dual-Encoder vs Cross-Encoder for Gene Regulatory Network Inference
Target: IEEE TNNLS
Parameter-matched (5.58M) comparison across four imbalance regimes. Ablation, pruning, and cold-start evaluation.
Submitted

What I work with

Languages
Rust Python TypeScript C++ React
Machine Learning & Deep Learning
PyTorch TensorFlow CUDA TensorRT MLflow Optuna MediaPipe OpenCV
Infrastructure & Tools
Docker FastAPI AWS wgpu Tauri Streamlit Gradio Weaviate
Research domains
Computational Biology Gene Regulatory Networks Clustering / Unsupervised ML Agent-Based Modelling Keynesian Macroeconomics Time Series / Finance

Let’s work together

Open to research collaborations, open-source ideas, or any dataset that made you stop scrolling.