Evint Leovonzko

Cofounder of
Kolosal AI

Making AI genuinely useful, at Kolosal AI and previously at Genta Technology. Researcher at UBC, where I implement and explore how machines decode biology (from circular RNA to tumor suppressor genes) and, occasionally, the stock market.

See my work About me
01 / Projects

Things I've built

Open-source tools, platforms, and applications

Open Source · Machine Learning

SOM Plus Clustering

Extended self-organizing map library with PCA-guided initialization, silhouette scoring, and Davies-Bouldin evaluation. Built for reproducible unsupervised clustering in scientific workflows.

Python NumPy SciPy scikit-learn
Learn More → Repo ↗
Open Source · Physics

Chain Reaction Simulation

High-performance nuclear chain reaction simulator in Rust. GPU-accelerated particle physics with real-time 3D visualization, 150+ isotope library from ENDF/B-VIII.0 nuclear data.

Rust wgpu GPU Nuclear Physics
Learn More → Repo ↗
Open Source · Finance

Faction: Stock Analysis App

Desktop app for evaluating stocks through technical and fundamental analysis, powered by on-device ML inference and a local Qwen2.5 LLM. It runs fully offline, with no cloud required.

Tauri Rust React LLM
Learn More → Repo ↗
Community · Education

Rebirth Educational Platform

Community-driven platform connecting learners and researchers through structured courses and peer mentorship programs.

Education Community
Learn More →
Open Source · Neuroscience

TRIBE v2: Neural Prediction Workstation

Interactive 3D brain visualization that predicts cortical activation across six regions in response to text, image, and audio stimuli. Powered by LLaMA, CLIP, and Wav2Vec with a Rust backend.

Rust Three.js LLaMA CLIP
Learn More → Read Paper → Repo ↗
Open Source · Quantitative Finance

orderflow-rs

High-performance Rust pipeline for limit order book microstructure research: OFI features, 87 technical indicators, synthetic LOB simulation, and walk-forward backtesting from a single binary.

Rust LOB OFI Backtesting
Learn More → Repo ↗
02 / Research & Analysis

Papers & analyses

First-author research in ML, computational biology, quantitative finance, and computational economics

Fiscal Stabilisers, Minsky Dynamics, and Distributional Outcomes in a Keynesian ABM
Computational Economics · Agent-Based Modelling
Keynesian ABM, zero lower bound, Minsky dynamics, fiscal policy, Rust, Monte Carlo
Daily-frequency, 10,000-agent closed-economy ABM reproducing four stylised macroeconomic facts. Novel result: ZLB binds in 93.7% of simulation periods (MC median), mirroring post-1998 Japan and post-2013 Euro-area, with a bimodal distribution driven by TFP-variance threshold effects.
First Author
Individual Optimality and Collective Failure in a Keynesian Agent-Based Model
Computational Economics · Coordination Failure
coordination failure, grid search, survivability optimisation, decentralised equilibrium, Keynesian ABM
Grid search identifies survival-maximising strategies for households, firms, and banks. The all-best economy (all types simultaneously following individually-optimal rules) produces GDP growth −65%, firm bankruptcies 17× higher, and negative mean firm profit: a structural coordination failure attractor.
First Author
SOM-TSK: Topology-Seeded Clustering Framework
Machine Learning · Clustering
self-organizing maps, K-means, Calinski-Harabasz, Rust
A clustering framework exploiting SOM topology to seed deterministic K-means: three novel seeding strategies, DenSOM density clustering, and AutoSOM k-selection. The result is 6 wins, 0 losses vs KMeans++ across 24 benchmark datasets.
First Author
GRASP: Graph-Routed Adaptive Spectral Partitioning
Machine Learning · Clustering
spectral clustering, GNG, Nyström, oracle routing, parameter-free
Autonomous topology-aware clustering pipeline: GNG topology → normalised spectral gap → oracle routing to 6 specialist algorithms. Parameter-free, processes 50K points in 140ms, ARI 0.378 on interlocking rings vs DBSCAN’s 0.055.
First Author
Dual-Encoder vs Cross-Encoder for GRN Links
Computational Biology · IEEE TNNLS
deep learning, GRN inference, encoder architecture
Parameter-matched comparison (5.58M params) of dual-encoder and cross-encoder architectures across four imbalance regimes, ablation, pruning, and cold-start evaluation. Submitted to IEEE TNNLS.
First Author IEEE TNNLS
Modular vs Monolithic GRN Architectures
Computational Biology · Research
cross-encoder, gradient failure, AUROC
Diagnosed three critical gradient failures in two-tower GRN models, then compared corrected modular and monolithic cross-encoders. The cross-encoder achieves AUROC 0.904 vs 0.810, with far greater imbalance robustness.
First Author
Two-Tower Networks for GRN Inference
Computational Biology · scRNA-seq
two-tower MLP, scRNA-seq, Rust
Pure-Rust two-tower MLP learning entity embeddings and cell-type expression profiles to predict TF–gene interactions: 83.06% ensemble accuracy, CPU-trainable, no deep learning framework.
First Author
Deep Learning for Circular RNA
Computational Biology · Published
circRNA, ANN, disease classification
Published ANN pipeline classifying circRNA-disease associations with Gaussian blur preprocessing: 75% accuracy at 0.14ms per prediction.
First Author
Functional Group Analysis
Cheminformatics · Research
GAT-VGAE, SOM, drug-likeness
GAT-VGAE + SOM pipeline analysing 249,455 ZINC15 molecules with counterfactual QED decomposition. It maps functional group enrichment across drug-like chemical space in Rust.
Cheminformatics
GDP Trajectory Clustering
Economics · Analysis
unsupervised clustering, macroeconomics, k-means
Unsupervised clustering of 190 countries over 45 years reveals four distinct economic paths, from stagnation to exponential growth.
Analysis
Crypto vs Stock Timing
Finance · Analysis
Bitcoin, cross-correlation, Granger causality
Do crypto markets lead or lag equities around recessions? Cross-correlation and Granger causality analysis across five business cycles.
Finance
Bitcoin Portfolio Allocation
Finance · Analysis
risk budgeting, BTC sizing, portfolio optimization
Optimal BTC sizing via Risk-Budget Framework. Component Risk Contribution analysis across five portfolio profiles finds 10–12% is the evidence-based ceiling.
Finance
What Does the Internet Do to the Brain?
Computational Neuroscience · Digital Media
fMRI encoding, internet content, cortical mapping, deep learning, Global Workspace Theory, TRIBE v2
Activation Cartography maps 3,008 stimuli across 13 content categories against TRIBE v2 predictions, a 177M-parameter fMRI encoder. Significant main effect F(12, 2995) = 13.51, p < 10²&sup6;; ThreatSafety drives broadest cortical activation; dominant gradient accounts for 96.9% of between-category variance.
First Author Neuroscience
Cancer Leading Mutation DNA of P-53 Gene
Computational Biology · Award-Winning Research
P-53, genetic algorithms, tumor suppressor, cancer genomics, mutation analysis
Closed-loop genetic algorithms surface the earliest mutation signatures that destabilise P-53, the “guardian of the genome”, before malignant cascades take hold. Best Research Project, UBC Vantage College.
First Author
1 / 3
The Map

How it all connects

Every project and paper, mapped to the domains they touch.

Try the live SOM playground →
03 / Background

Experience

Current

University of British Columbia

Undergraduate Researcher

Building deep learning pipelines for circular RNA classification and genomic sequence analysis.

2023 to Present

Kolosal AI

Co-Founder

Architecting open-source tools for local LLM deployment, inference optimization, and MLOps automation.

2024

Publication

First Author

"Deep Learning Algorithm with Gaussian Blur Data Pre-processing in Circular RNA Classification"

04 / Who I am

About

I chase problems where data hides something meaningful: a motif buried in RNA sequences, a leading signal in financial time series, or a chance to make ML infrastructure less painful.

Currently an undergraduate researcher at UBC building deep learning systems for genomic data, and co-founding Kolosal AI to make running LLMs locally actually simple. My first-author publication on circular RNA classification is live in URNCST Journal.

When I'm not debugging loss curves, I'm reading about macroeconomics or exploring Vancouver's trails.

More about me

What I work with

  • Python
  • PyTorch
  • TensorFlow
  • FastAPI
  • Docker
  • CUDA
  • Pandas
  • OpenCV
  • Streamlit
  • MLflow
Open to opportunities

Let's build
something together

Always open to interesting conversations, whether it's a research collaboration, an open-source idea, or a dataset that made you stop scrolling.

LinkedIn GitHub