Rigor before scale
Correctness and reproducibility come first. Scaling a wrong result just gets there faster.
Contents
TABLE OF CONTENTS
AI & Distributed Systems Researcher / Engineer
Costin
Alexandru
Deonise
PhD-track researcher at POLITEHNICA Bucharest, CTO building production AI systems, and national coach for Romania’s AI Olympiad team.
Chief Technology Officer
Leading AI system delivery for enterprise clients at Qflex Technologies.
PhD-track Researcher
Parallel & Distributed Computer Systems, POLITEHNICA Bucharest.
National Team Coach, IOAI
Coached Romania's AI Olympiad delegation to a 4th-place worldwide finish in 2026.
6 Publications
On scalable differentiation, privacy-preserving ML, and LLM systems.
Associate Lecturer
1000+ students taught at POLITEHNICA Bucharest since 2022.
Robotics Mentor
Lead mentor for AlphaZ FRC #11141, a FIRST World Championship award winner.
Selected work
Recent, high-signal results
Speaker-attributed meeting transcription refinement with constrained open-weight language models
Future Generation Computer Systems (Elsevier), June 2026
Homomorphic Encryption for Privacy-Preserving Machine Learning Inference
Annals of the Academy of Romanian Scientists, Series on Science and Technology and Information, Vol. 18, No. 2, 2025
Best Scientific Paper · Most Impactful Paper
Stochastic Taylor Derivative Estimator for High-Order Differential Operators
UPB Students' Scientific Communications Session (SCSS 2025), 2025
2026
Adobe Systems Romania Scholarship
2025
Most Impactful Scientific Paper Award · Engineering Domain
2025
Best Scientific Master's Paper Award · Engineering Domain
2025
"Teodor Ceaușu" Scholarship for Excellence in Research and Technological Development
Site structure
Explore
Research
Scalable differentiation, privacy-preserving ML, and LLM systems.
Publications
Six peer-reviewed papers across journals and conferences.
Experience
From ML internships to Chief Technology Officer.
Teaching & Olympiads
Lecturing at POLITEHNICA and coaching Team Romania's AI Olympiad squad.
Projects
AlgoTrack, AtlasRAG, distributed STDE, and AlphaZ robotics.
Awards
Scholarships, paper awards, and competition results.
How I work
Three things I don’t compromise on.
Publish, then apply
Research earns its keep when it ships: from PDE solvers to production RAG pipelines.
Explain clearly
A concept isn’t understood until it survives being taught to 1000 students.
Trajectory
Researcher, engineer, teacher, coach.
01 / Research
Systems
That
Scale
From distributed high-order differentiation on multi-GPU clusters to privacy-preserving inference: published, peer-reviewed research, not slideware.
02 / Engineering
Production
AI
Systems
CTO at Qflex Technologies, leading AI system architecture and delivery for enterprise clients.
Built production RAG, knowledge-graph extraction, and speech pipelines at the Research Institute and Bitdefender.
Built AlgoTrack end to end, a tutoring platform deployed and in daily use.
03 / Teaching
1000+
Students
Taught
Associate Lecturer at POLITEHNICA Bucharest since 2022, across programming, parallel algorithms, and numerical methods.
