PhD Applicant · Dhaka, Bangladesh
Distributed Systems · Databases & Data Management · Machine Learning & AI Systems · Computer Systems
I hold a B.Sc. in Mechanical Engineering from BUET, completed in June 2026. My work sits where systems engineering meets computation. I held two research appointments there: an undergraduate thesis on hard constraint physics informed neural networks for 3D Navier-Stokes reconstruction, and a research internship building the data and deployment infrastructure for a medical imaging pipeline that ran in the field. Alongside those I worked in production on backend services decoupled through queues and scaled horizontally.
Across three IEEE publications I was responsible for the full implementation: data pipeline, model architecture, and evaluation. I am seeking to ground that practice in graduate research on distributed storage, consensus and systems reliability, and on the systems side of machine learning and AI: training and inference pipelines, serving on the device, and benchmarking anyone can reproduce. A secondary interest is high performance and scientific computing.
Where my systems engineering and my applied deep learning publications meet, and what backs each.
Consensus, replication, fault tolerance
My primary interest. I have built and operated services where correctness depends on what survives a crash: horizontally scaled workers on Redis Streams and BullMQ, stateless stages, idempotent retries. I want to study the formal side of what I so far know only operationally.
Storage engines, query execution, indexing
Taalaash raised these questions before I had the vocabulary for them: how to index 20,000+ pages of OCR output cheaply, when a vector index beats an inverted one, where caching belongs at 100,000+ users. I am working through CMU 15-445 and BusTub to replace intuition with mechanics.
The infrastructure around a model, not only the model
Every one of my publications needed a working system before it produced a number: reproducible pipelines, cross validated benchmark harnesses, evaluation anyone could run again. I am most interested in the serving layer, in inference that runs on the device and the memory and latency budget it has to fit.
Learned surrogates against classical solvers
My thesis reconstructs 3D Navier-Stokes fields for urban flows at high Reynolds number using hard constraint PINNs, benchmarked against ANSYS Fluent. The question that interests me is the trade: what a surrogate buys in inference time and what it costs in guarantees.
API design, deployment, operations
Four years of production work: API design, containerised deployment, CI/CD, and the unglamorous operational tail. It is the layer that makes research reproducible by someone other than its author.
Three IEEE conference papers. I wrote the full implementation of each: pipeline, model, evaluation.
Z. I. M. Rayat, M. N. M. Monim, M. Mamun · IEEE QPAIN 2026
Contribution. First author. Designed the weather aware variable attention mechanism and built the full forecasting pipeline over eight years of meteorological data, reaching R² = 0.9956 on hourly prediction.
M. N. M. Monim, Z. I. M. Rayat, M. Mamun · IEEE QPAIN 2026
Contribution. Implemented the hard constraint PINN formulation and the curriculum learning schedule in full, in TensorFlow and DeepXDE, then benchmarked inference against ANSYS Fluent for physical accuracy.
M. N. M. Monim, Z. I. M. Rayat, S. Javed · IEEE ICTP 2025
Contribution. Built the real time speech to sign pipeline (ASR, a landmark based MLP classifier, and servo control), reaching 91.11% test accuracy across 38 gesture classes on a 3D printed robotic hand.
A scabies and non-scabies image dataset verified by dermatologists in Bangladesh, paired with a transfer learning benchmark across seven ImageNet pretrained CNN backbones under five fold stratified cross validation, Grad-CAM and Eigen-CAM explainability, and retrieval analysis over the learned embedding space.
Undergraduate Thesis
Bangladesh University of Engineering and Technology · Department of Mechanical Engineering · Advisor: Dr. Mohammad Mamun
Oct 2024 to May 2025
Institute of Appropriate Technology (IAT), BUET · Cyber Physics Intelligence Laboratory · Advisor: Dr. Tahsina Farah Sanam
May 2025 to Sept 2025
RubizCode IT
Research artifacts and production systems, grouped by how close they sit to the interests above.
A Node.js API enqueues jobs to Redis (BullMQ). Python workers that scale horizontally pull those jobs, extract and OCR documents with PyMuPDF and Tesseract, then hand structured output to an embedding queue through S3 compatible object storage. Extraction and embedding scale independently and share no state.
A search engine over Bangla academic material, indexing 20,000+ textbook pages through a Tesseract pipeline tuned for Bangla, with AI reranking over vector embeddings and caching at both the CDN and the data layer.
A fully offline desktop app running Llama 3.2 locally through mistral.rs, with no cloud dependency and no telemetry. The inference backend is selectable at compile time, either linked in process or loaded as a DLL, across x64 and arm64 builds.
Distributed video processing on Go and Redis Streams, handling partial failure and consumer group recovery across FFmpeg transcoding workers.
Streaming from several cameras at once over WebSockets with OpenCV processing, at low enough latency to drive a Mars rover. Built for Team Interplanetar.
A weather aware variable attention mechanism over eight years of meteorological data, forecasting Dhaka's hourly air quality. First author on the IEEE QPAIN 2026 paper.
A React Native app pairing an image classifier that runs on the phone with a symptom questionnaire, to screen for scabies lesions. Deployed for a dormitory field pilot at IAT BUET.
A real time pipeline translating spoken Bangla into sign gestures on a 3D printed robotic hand: ASR, a landmark based MLP classifier, and servo control. IEEE ICTP 2025.
A full RAG pipeline for chunking, embedding, retrieval and reranking over textbook sources, serving an AI tutor behind a real time NestJS backend.
The implementation behind the thesis and its QPAIN 2026 paper: hard constraint PINNs with curriculum learning, reconstructing 3D velocity and pressure fields and benchmarked against ANSYS Fluent.
Bangladesh University of Engineering and Technology (BUET) · CGPA 3.10 / 4.00
Notre Dame College, Dhaka · GPA 5.00 / 5.00
Linear Algebra & Matrices · Numerical Analysis · Differential & Integral Calculus · Ordinary & Partial Differential Equations · Vector Calculus · Complex Variables & Harmonic Analysis · Laplace Transformation & Series Solution · Computer Programming Language & Lab · Electro-Mechanical System Design Lab
CS50x 2026, Harvard University (OpenCourseWare) · Algorithms Part I & II, Princeton University (Coursera) · Algorithms Specialization, Stanford University (Coursera) · Algorithmic problem solving in C++ (LeetCode, NeetCode.io)
MIT 6.5840 Distributed Systems · CMU 15-445 Database Systems
Finalist, Build4Democracy Hackathon
United Nations Development Programme (UNDP)
Finalist, Innovation Founder (UIHP Cohort)
RISE
Champion, Speak Out For Engineers
Institution of Mechanical Engineers (IMechE)
Technical Scholarship
Bangladesh University of Engineering and Technology
Networking in Google Cloud: Fundamentals
President · BUET Animata, Animation & VFX Club
Since Mar 2025Founding Assistant General Secretary. Organised workshops on 3D modelling and animation, and designed the club's visual brand identity.
Web Master · Team Interplanetar, BUET Mars Rover Team
Oct 2022 to Oct 2024Built and maintained the team's official website and a certification system for Mars rover engineering achievements.
Executive · BUET Robotics Society
Jul 2022 to Jul 2023Built and maintained the society's web presence and internal certification tooling.
Happy to discuss research fit, share code or data from any of the work above, or send a transcript.
Last updated August 2026