CV

PhD Candidate

Algorithmics group, EEMCS, TU Delft

Researching how optimization solvers justify their answers, and how they can be made to reason about the global structure of a problem instead of one constraint at a time.

• Developed a branch-and-bound method for finding the shortest resolution proofs of unsatisfiability, and showed that shorter proofs are not tidied-up versions of long ones: across the SAT Competition benchmarks, proofs that trim well almost never shrink further, and proofs that shrink well almost never trimmed well.
• Introduced a preprocessing method that infers new Cumulative constraints for scheduling problems by reading such constraints as linear inequalities, discovering covers of tasks that cannot run in parallel, and strengthening them by lifting. This approach found 25 new lower bounds and five new best solutions on standard RCPSP benchmarks.
• Designed a propagator that aggregates disjointness information across several constraints to find cliques of mutually disjoint tasks — closing instances that a solver reasoning over each Cumulative constraint separately provably needs exponential time to refute.
• Contributed to the proof logging toolchain for constraint programming: a multi-stage framework that keeps the runtime cost of emitting certificates typically below 10%, pseudo-Boolean reasoning that certifies dynamic programming and decision diagram algorithms, and a human-auditable proof system for constrained shortest paths.
• Supervised MSc and BSc thesis projects in constraint programming, search methods, and combinatorial optimization, aiming for publishable research.
• Organized a research seminar about combinatorial optimization.

Senior Data Scientist

X5 Group

Participated in the development of the analytics platform for the transport division of X5 Group, the largest Russian retailer.

• Refactored the report generation engine, which eliminated multiple inconsistencies in the resulting reports and improved the loading time by an order of magnitude. The updated engine also resulted in a simpler source code, which allowed the development team to expedite the shipment of the new features.
• Designed a probabilistic model of the relation between the vehicle repairs and their mileages. This model laid the foundation for the demo widget displaying the planned repair budget.
• Developed a statistical model for vehicle fuel usage during its trip and prototyped a widget for detecting abnormal behavior of vehicles and their drivers.
• Improved the loading speed of the dashboard by 15 times
• Halved the application data pipeline runtime while correcting multiple data consistency errors

Algorithmic Software Engineer

Adeptik Plus OOO

Designed algorithms for applied combinatorial optimization in various domains, including last-mile logistics and production planning.

• Automated the dispatching of field employees for ICL Services, which helped to reduce the time to reaction by 4 times and push the SLA compliance rate to 99.5%
• Developed a system for scheduling deliveries for a local bakery, halving their vehicle maintenance costs

Master's degree, Mathematics

Moscow Institute of Physics and Technology

Contemporary Combinatorics

4.81 / 5.00, with distinction

• Random Graphs
• Additive Combinatorics
• Analytic Combinatorics

Bachelor's degree, Software Engineering

Astrakhan State University

4.95 / 5.00, with distinction

Yandex School of Data Analysis

Big Data track

4.92 / 5.00

• Convex Optimization
• Bayesian Methods in Machine Learning
• Advanced Deep Learning

Paths, Proofs, and Perfection: Developing a Human-Interpretable Proof System for Constrained Shortest Paths

Proceedings of the AAAI Conference on Artificial Intelligence

We have developed an approach for deriving compact tree-like optimality proofs for constrained shortest path problems. Our approach has been able to derive proofs comparable in width with the MIP proofs, and consistenly narrower proofs for some instance classes.