PhD Candidate · Applied Mathematics · UCF · Orlando, FL

Abdulla Toleugazin

I build statistically rigorous methods for messy, structured data: spectral algorithms for multilayer networks with provable guarantees, and validation workflows that test what a metric or a trading signal really measures before anyone relies on it.

Email me GitHub LinkedIn Résumé PDF ↗

Focus: Multilayer networks
Talks: NetSci 2026 · NYS GMC
Education: UCF · MIPT
Open to: Quant research · AI/ML research

About

I'm a PhD candidate in Applied Mathematics at the University of Central Florida, advised by Prof. Marianna Pensky, working on network science, graph theory and statistics. My dissertation develops spectral methods for multilayer networks: algorithms that find shared latent structure across heterogeneous layers and use it to classify weaker ones, backed by non-asymptotic theoretical guarantees.

Before UCF I studied Applied Mathematics and Physics at the Moscow Institute of Physics and Technology (MIPT). Alongside research I've worked as an ML evaluation intern at Handshake MOVE, stress-testing LLMs on olympiad-level mathematics, and as a data/identity analyst intern at Black Mile Strategy, tracing customer identities across legacy systems for a cloud migration.

Across projects I care about the same thing: separating what the data supports from what it only appears to show. That means pre-registered tests, honest out-of-sample checks, and reporting what didn't work. I'm looking for quantitative research roles in finance/trading and research roles in AI/ML, statistics and optimization.

Research & Papers

In preparation Multilayer networks Spectral methods Semi-supervised learning

Semi-supervised Classification in Rectangular Multilayer Networks with Groupwise Subspace Structure

Abdulla Toleugazin, Marianna Pensky · Department of Mathematics, UCF

Studies a heterogeneous rectangular multilayer network model where layers record interactions between a common set of row nodes and layer-specific column nodes. Introduces a two-stage algorithm (automatic identification of active layers via hollowed Gram embeddings, followed by group-level subspace aggregation) with non-asymptotic guarantees for perfect layer clustering and consistent subspace estimation.

Experiments code ↗ arXiv preprint coming soon

Working paper Sports analytics Measurement validity Reproducible research

Stable Rankings, Uncertain Attribution: Validating NBA Matchup Turnover Metrics

Abdulla Toleugazin

A stable leaderboard is not proof that a metric credits the right player. Across eight NBA seasons of public matchup data, season totals reconcile with play-by-play within 1.5%, yet the recorded defender matches the officially credited player in only 30.9% of 8,560 checkable turnovers. A three-outcome decomposition separates engagement from disruption, and the paper turns these checks into a reusable workflow for auditing what a sports metric measures before it informs decisions.

Paper & replication code ↗

Featured Work

All repositories →

Dissertation research

Spectral Clustering for Multilayer Networks

Algorithms that find shared latent structure across heterogeneous network layers, cluster the informative (active) layers, and use them to classify weaker ones, with non-asymptotic guarantees.

Code ↗

160-layer MovieLens stress test: adjusted Rand index 0.371 vs. 0.178 for the next-best baseline and 0.062 for all-layer spectral clustering
Simulations with few wide and many narrow layers: roughly 3–5× lower clustering/classification error than all-layer refinement
Presented at NetSci 2026 and the 51st NY State Regional Graduate Mathematics Conference

Spectral methodsNetwork scienceSemi-supervised learningMATLAB

Gator Quant Hacks 2026 · Systematic Trading track

8-K Event-Driven Systematic Trading

A market-neutral strategy that shorts companies after they file acquisition-agreement 8-Ks, found by a pre-registered, multiple-testing-controlled search over 165 event configurations.

Code & quant note ↗

Out-of-sample Sharpe 0.82 net of costs and borrow, chosen on in-sample data only and run once
Random-date placebo beaten 100 of 100 times; beta to the S&P 500 ≈ 0
32 pre-registered hypotheses reported, including every failure, with a nine-pitfall statistical audit

PythonEvent studiesMultiple testingBacktestingMassive API

Knight Hacks IX · BNY Financial Services & Databricks challenges

CutoverGuard

A record-level cutover gate for financial data migrations on Databricks. It passes good migrations and blocks bad ones, including wrong-owner errors that every totals check misses.

Code ↗

Verified on 7.5M orders: caught all 811 wrong-owner orders while totals stayed identical
15-table Lakeflow bronze → silver → gold pipeline ending in a single BLOCKED / READY decision
The failing gate stops promotion automatically; the corrected migration promotes all 7.5M rows

DatabricksLakeflowSQLPythonData quality

More projects

BlueSky NetSci Hackathon ↗

Spectral clustering and Louvain community detection on a starter-pack network reaching 1.8M accounts and 10.6M memberships.

Medlaunch Graph Challenge ↗

Python/PyMongo graph layer over healthcare quality data: reference validation, multi-hop queries, similarity edges, connected components.

FedProx-Share ↗

Federated learning algorithm extending FedProx.

DIMPLE ↗

Directed research with Prof. Marianna Pensky.

Experience & Education

University of Central Florida

Aug 2022 – Present

Graduate Research Assistant · Orlando, FL

PhD research with Prof. Marianna Pensky on clustering and classification in multilayer networks.
Developed a two-stage method that clusters informative layers, estimates their shared latent subspaces, and classifies weaker layers, combining spectral methods with statistical theory.
In a 160-layer MovieLens stress test, reached an adjusted Rand index of 0.371 vs. 0.178 for the next-best tested baseline and 0.062 for all-layer spectral clustering.
In simulations with a few wide, informative layers and many narrow ones, reduced clustering/classification error by roughly 3–5× versus all-layer refinement.
Established non-asymptotic guarantees for exact layer clustering and consistent subspace estimation under stated assumptions.

Black Mile Strategy

Jun 2026 – Sep 2026

Data / Identity Analyst Intern

Supported a national retailer's customer-data modernization: mapped customer identities and traced transaction flows across legacy MDM, CDB, CDM and mainframe systems ahead of a cloud migration.
Translated SAS logic for ZIP-code-based location assignment into Python.
Produced data-flow documentation and identity-mapping rules used to guide the migration.

Handshake MOVE

Apr 2025 – Apr 2026

ML Evaluation Intern · Remote

Used Python, pandas, NumPy and scikit-learn to organize and validate evaluation data for 500+ advanced mathematical reasoning tasks at IMO/HMMT level.
Reviewed LLM-generated solutions, documented recurring logical errors and annotation gaps, and designed adversarial test cases to refine evaluation criteria.

University of Central Florida

Aug 2022 – Present

Mathematics Instructor / Teaching Assistant

Instructor/TA for Intermediate Algebra, College Algebra, Precalculus and Trigonometry.
Led three Calculus I recitations (about 150 students); taught Calculus II and ODE sections (about 50 students).
Supported courses with 300–1,500 total enrollment across lectures, office hours, quizzes, exams and grading.

Education

Ph.D., Applied Mathematics

University of Central Florida

Aug 2022 – Aug 2027 (expected) · Advisor: Prof. Marianna Pensky

M.S., Applied Mathematics

University of Central Florida

Aug 2022 – Aug 2024

B.S., Applied Mathematics and Physics

Moscow Institute of Physics and Technology (MIPT)

Sep 2018 – Jun 2022 · Moscow, Russia

Talks & Conferences

Jun 2026

Semi-supervised Classification in Rectangular Multilayer Networks with Groupwise Subspace Structure

NetSci 2026

June 3, 2026

Mar 2026

Semi-supervised Classification in Undirected Multilayer Networks with Groupwise Subspace Structure

51st Annual New York State Regional Graduate Mathematics Conference

Syracuse University · March 21, 2026

Skills

Languages & Libraries

PythonpandasNumPySciPyscikit-learnPyMongoMATLABRSQLLaTeX

Methods

Spectral methodsStatistical theoryClustering & classificationNetwork scienceGraph theoryML evaluationTree-based methodsMultiple-testing control

Quant & Finance

BacktestingEvent studiesFactor attributionOptions pricing (Black–Scholes)

Data & Compute

Databricks / LakeflowMongoDBETL & data qualityGitJupyterMatplotlibExcelCPU/GPU on UCF ARCC cluster