Education
Taras Shevchenko National University of Kyiv
Bachelor of Computer Science
GPA: 3.91/4.00
Kyiv, Ukraine
Sep 2022 - Jun 2026
Relevant Coursework: Algorithms & Data Structures, Machine Learning, Neural Networks, Probability & Statistics, Distributed & Parallel Programming, Computational Geometry, Cryptography
Publications & Submissions
NaviLoc: Trajectory-Level Visual Localization for GNSS-Denied UAV Navigation
Jan 2026Pavel Shpagin and Taras Panchenko
Drones (MDPI), 2026, 10(2), 97 | Published
DOIPractical Polygonal Triangulation in O(n + r log r) Time
Nov 2025Pavel Shpagin and Vasyl Tereshchenko
Computational Geometry: Theory and Applications | Under review
DOIMultiplication-Free O(b log b) Balanced Numeral Conversion
Dec 2025Pavel Shpagin
Information Processing Letters | Under review
DOITriangulating Simple Polygons: A Survey
Nov 2025Pavel Shpagin
Journal of Computational Geometry | Under review
DOIExperience
Academia Techtheacademia.tech
Co-founder & Research Scientist
Kyiv, Ukraine
September 2025 - Present
- >Developed NaviLoc, a GNSS-denied UAV localization system; published in Drones (MDPI), a Q1 journal.
- >Optimized navigation algorithms 10x, making Pi 5 deployment feasible.
- >Accelerated server preprocessing by 20x, enabling faster missions at scale.
- >Built and deployed a novel agentic Signal SupportBot, serving chats with 1000+ users and 2000+ messages/month.
Grammarly
Software Engineer Intern
Kyiv, Ukraine
June 2025 - September 2025
- >Invented a new ensemble to reduce the costs of feedback evaluation by 10x, while increasing accuracy by 30%.
- >Developed a novel lightweight agent to cut preprocessing costs by 20x and boost data quality by 10%.
- >Led annotations on 900 multilingual data, to build gold datasets for model evaluation.
- >Processed 3 Million user data with LLM-based methods, used to train and deploy a model to millions of users.
Pilgway LLC
AI Engineer
Kyiv, Ukraine
December 2024 - June 2025
- >Designed new GPS-denied localization algorithms that boost localization accuracy from 600m to 20m, yielding an effective 30x improvement for long trajectories.
- >Built full and robust server and UAV infrastructure from scratch.
Projects
Library+
Azure, Jira, OpenAI API, Figma
- -Led a team of 7 people to build a mobile app for the government.
- -Designed and architected the full app system.
- -Deployed AI chatbot, search engine, and core product.
Casters
NextJS, Huggingface, Pytorch, pandas
- -Created a full-stack NextJS web application with an AI card generator.
- -Preprocessed datasets for training a 12B model.
- -Fine-tuned and deployed a VLM model using Huggingface.
Ai ArtBase
FastAPI, React, ChromaDB, PostgreSQL, Docker
- -Built a gallery website with multimodal RAG search engine.
- -Scraped over ~30K AI images across different galleries.
- -Implemented an advanced gallery navigation algorithm with complex route caching and restoration.
Parallel Kernel SVM
C++, MPI, OpenMP
- -Developed a parallelized kernel SVM algorithm based on quadratic programming.
- -Trained the model on 10K Amazon review records.
- -Achieved 3x speedup over baseline implementations, maintaining 80% accuracy.
CUDA Matrix Multiplication
CUDA
- -Built a matrix multiplication algorithm that's 10x faster than the baseline.
- -Increased memory throughput by 8x, using advanced optimization techniques.
- -Reduced energy consumption of the optimized algorithm by 5x.
Technical Skills
Awards & Honors
Brave1 Research Grant
NaviX
Awarded $80,000 research grant from Brave1 for developing advanced UAV visual localization systems.
ICPC Ukraine 2025
Programming Contest
Ranked in the top 12% nationwide among university teams.
Meta Hacker Cup 2025
Programming Competition
Advanced to Round 2.
IEEEXtreme 19.0 2025
Global 24-hour Programming Contest
Competed as team IEEExceptional; placed around top ~25% in a contest with 18,500+ registered participants.
Certificates
Linear Algebra for Data Science
MotivateYourself | Instructor: Ian Tsybulkin
Neural Networks and Computer Vision
Stepik | Score: 87%