Руслан

Ruslan Murha
Computer Science Student | Risk & Data Analytics
Warsaw, Poland · [відкрити контакти](див. вище в блоці «контактна інформація») · [відкрити контакти](див. вище в блоці «контактна інформація») · [відкрити контакти](див. вище в блоці «контактна інформація») · github.com/mvrhaa

PROFILE
Second-year Computer Science student at PJATK, building Python and SQL tools for bank risk data: parsing Pillar 3
disclosures and modelling credit risk capital. Winner of a quantum optimization hackathon track (Oct 2026). Seeking a risk
or data analytics internship. Available 25–30 h/week (hybrid or on-site in Warsaw) and full-time in summer.

EDUCATION
Polish-Japanese Academy of Information Technology (PJATK) Oct 2025 – February 2029
BSc in Computer Science, 2nd year, Warsaw
Coursework: Linear Algebra, Mathematical Analysis, Discrete Mathematics, Java & OOP; ongoing: Statistics & Probability,
Relational Databases (SQL), Algorithms & Data Structures.

ACHIEVEMENTS
1st Place, Quantum Optimization Track — IBM Qiskit Fall Fest Hackathon Warsaw, Oct 2026
Led a team of 5 and presented our solution to the jury. Implemented QAOA, ran the benchmarks against classical methods
and built the Streamlit demo (see LODQA in Projects).

PROJECTS
PillarScope — Pillar 3 disclosure checker · Python, pandas, LLM API, Pydantic, pytest
• Extracted CET1, Tier 1, leverage, LCR and NSFR from the EU KM1 tables of Polish banks’ Pillar 3 reports.
• Reconciled extracted capital ratios with EBA Transparency Exercise data at a 0.10 pp tolerance; mismatches were routed
to manual review instead of being accepted. LCR and NSFR were checked against their own components
(HQLA/outflows, ASF/RSF).
• Validated every LLM answer with Pydantic (plausible ranges, units, ratio order) and retried rejected answers; covered
parsing and validation rules with 79 pytest tests.
MoCulator — PD model with Margin of Conservatism · Python, scikit-learn, OpenPyXL, Excel
• Estimated PD on a synthetic portfolio of 5,000 loans with 5 years of default history using logistic regression (out-of-time
AUC 0.77).
• Applied Margin of Conservatism (categories A–C, EBA GL on PD/LGD estimation) and computed RWA with the
Foundation IRB risk-weight formula under CRR3; MoC raised RWA by 15% and lowered CET1 by 2.2 pp.
• Generated an Excel report with live formulas (55,000+ formulas, results match Python) so every figure can be traced and
recalculated by hand.
LODQA — Constrained optimization with QAOA · Python, Qiskit, Streamlit
• Formulated level-of-detail selection for 3D scenes under a polygon budget as a knapsack-type QUBO on 19 qubits (same
problem class as budget-constrained portfolio selection), verified by checking all 524,288 (2^19) states against the exact
solution.
• Solved it with QAOA (X and XY mixers, depth 1–4) and compared against exact dynamic programming, a greedy baseline
and random sampling on 36 runs (3 hard scenes × 3 seeds × 4 depths): the XY mixer sampled the exact optimum in all 36
runs, while greedy stopped at 86–87% of it.
• Built a Streamlit demo with a 3D viewer; the best setting (XY mixer, CVaR training) found the optimum in 1 of 9 shots,
28× more often than a random valid plan. Team of 5, 1st place.

SKILLS
Risk & regulation (project-level): Basel III / CRR capital and liquidity ratios (CET1, LCR, NSFR), Pillar 3 disclosures, PD
estimation, Margin of Conservatism, RWA
Data: Python (pandas, NumPy, SciPy, Matplotlib, scikit-learn), SQL (PostgreSQL, MySQL), Excel (formulas, OpenPyXL
automation)
Engineering: pytest, Pydantic, FastAPI, Git, Bash, Java

LANGUAGES
Ukrainian (native), Russian (native), English (C1), Polish (B2), German (B1)