Doctor · Researcher · Engineer · Educator

Dr. Vahram Martirosyan

Interpretable ML for Ocean Science · Explainable AI · Edge AI · HPC & Compiler Engineering

Doctor of Computational Mathematics with 23+ years across research, industry, and academia. I build intrinsically interpretable polynomial regression models for ocean and climate science, medicine, materials, and remote sensing — and the compilers and runtimes that make them fast, from LLVM auto-parallelisation to sub-millisecond inference on tiny devices. Currently Chief Scientist at ExplainableAI.app and Assistant Professor at Higher Colleges of Technology, Abu Dhabi.

ExplainableAI.app ↗ Projects
Dr. Vahram Martirosyan
23+
Years of experience
~20
Years of teaching
25+
Publications
Track record

Significant achievements

Headline results from research, books, and teaching — on-device inference, certified design, GPU speed-ups, and published textbooks.

958×
peak GPU speed-upLEAP · ITCA 2026
0.88 ms
on-device inference (ESP32)eICocean 2027 Forthcoming
R² 0.973
ocean temperature reconstructioneICocean 2027 Forthcoming
16.7%
certified CO₂ saving, concrete designICACTE 2027 Forthcoming
3 books
Operating systems & Linux programming2011, 2022, 2024
See all achievements →
What I work on

Focus areas

Making complex computational problems both efficient and interpretable — no black boxes.

Interpretable ML for Ocean & Climate

Polynomial zone ensembles for global ocean temperature, salinity, and dissolved-oxygen reconstruction, and significant-wave-height modelling.

Ocean T/S/DO WOA / IAP Sub-MB models

Explainable AI

Polynomial regression models with stagewise and elastic-net regularisation, explicit feature importance, and model sparsification.

PRMs Interpretability Compression

Edge / On-Device AI

int8/int16 quantisation and fixed-point kernels running sub-MB models on ESP32, RP2350, and Arduino with strict latency and energy budgets.

TinyML ESP32 180 µJ/inference

HPC & Compiler Engineering

LLVM/Clang passes for automatic parallelisation and binary lifting, with CPU and GPU (OpenCL/SPIR-V) backends — no source modification required.

HPC++ LEAP LLVM

Systems Programming

C/C++ expert with Linux kernel modules, device drivers (Ext4, XFS, JFS, BtrFS), Win32 API, GTK+, and embedded/IoT experience.

Linux kernel Drivers IoT

Teaching & Academic Leadership

Assistant Professor, designing and standardising courses across campuses in programming, architecture, IoT, and statistics.

HCT Abu Dhabi Curriculum Mentoring
Selected work

Projects & demos

A mix of research platforms, engineering tools, and interactive browser demos.

Live

ExplainableAI.app

No-code ML platform built entirely on intrinsically interpretable polynomial regression models. Browser-side WebAssembly training, a hand-written C++ inference engine, and an AI App Builder.

Visit site ↗
Live demo

Client-Side Digit Recognition

An ensemble of 10 sparse polynomial networks recognising handwritten digits ~5 ms per prediction, running fully in the browser with no server.

Try the demo →
Live demo

Math Expression Editor

A visual editor for building and rendering math expressions with fractions, powers, and roots — LaTeX export, JPEG round-trip, and offline support.

Open the editor →
Report

HPC++ / LEAP Speed-up Report

LLVM-based automatic parallelisation with heterogeneous CPU-GPU execution: 8.4× average and up to 958× peak GPU speed-up, plus byte-identical lifting of legacy x86-64 binaries.

Read the PDF ↗
Books & teaching

Books, tutorials & school

Beyond research: textbooks, teaching materials, and building a programming school.

Books

Three books on operating systems, systems programming, and Linux fundamentals.

All books →

Linux programming tutorials

Co-authored two Russian-language tutorials on Linux/POSIX programming, from system calls to the GUI, used at MIPT and RAU.

MIPT 2011RAU 2010
See in publications →

Founder, MutEx School

Founded and ran a programming school in Yerevan (2019–2023), delivering courses and public lectures.

Yerevan2019–2023
See in CV →
Research

Selected publications

25+ publications spanning ocean and climate data science, medical ML, materials, and compiler engineering.

Accuracy and Interpretability from a Single Solver: Polynomial Regression up to Degree 5 for Concrete Compressive Strength Prediction

Accepted
ICACTE 2026 · Martirosyan, V.
R² 0.859 · 74% of retained terms match known concrete physics

On-Device TinyML for Multi-Variable Ocean Profiling

Accepted
eICocean 2027Forthcoming · Martirosyan, V.
R² 0.973 T / 0.794 S / 0.895 DO · 0.88 ms on ESP32 · 180 µJ/inference

Efficient Multi-Class ECG Classification Using Polynomial Regression with an Analytical Solver

Under Review
Applied Computational Intelligence and Soft Computing · Martirosyan, V.
82.75% six-class accuracy · 12 µs inference · beats CNN, RF, XGBoost, SVM

LEAP: Lifting Legacy x86-64 Binaries to Recompilable IR for Automatic Parallelization

Accepted
ITCA 2026, GuangzhouForthcoming · Martirosyan, V.
100%/98% liftability (43 programs) · 10.4× over -O0 · 958× GPU · 377/377 tests

High accuracy, no signal: an identifiability and external-validation audit of climate and nitrogen response in China's cereal yield data

Submitted
Computers and Electronics in Agriculture · Martirosyan, V.
5,316-row China maize–rice–wheat panel · level R² 0.50–0.92 · anomaly R² mostly −0.09 to 0.08

Interpretable polynomial surrogates for halide-perovskite screening: domain-cleaned bandgaps, ideal tandem limits, and the limits of forward prediction

Submitted
Solar Energy Materials and Solar Cells · Martirosyan, V. & N. Martirosyan
31,618 perovskites · 21.0 meV MAE / R² 0.794 · best ideal tandem 43.73%
Full list of publications →
Get in touch

Contact

Open to research collaborations, consulting, and talks.

Emailvmartirosyan@gmail.com LinkedInin/vahrammartirosyan ResearchGateVahram-Martirosyan-2 GitHub@vmartirosyan