I'm a CSE undergraduate at United International University in Dhaka, Bangladesh. I care about the ideas and research behind the models, not only how to run them. So I read the papers, build tools I can run and inspect, and write up what I learn so another student can follow it.
Now: reviewing papers and building an NLP tool · Latest: released Research Radar 1.0.0 (Nov 2026)
Research Radar maps a research field from its conference papers. It runs on a laptop CPU with no API keys and shows its results in a local web dashboard.
- Builds a corpus from 22 conference venues and arXiv through OpenAlex, Semantic Scholar and OpenReview, merging duplicates across sources.
- Groups papers into topics with TF-IDF, SVD and k-means, measures which topics are growing, and pulls open-problem sentences out of abstracts.
- Scores a new idea for novelty against the corpus. A scheduled GitHub Actions workflow keeps the data up to date.
Python · FastAPI · scikit-learn · SQLite · JavaScript
Explore Research Radar → · Built on RTanalysis by Faysal Mahmud: the core approach is his, and I use some of his code with his permission.
See the real dashboard
From a test run on 24,192 ICML and ICLR papers (2024 to 2026): 63 topics and 4,884 open-problem sentences in just over a minute on a laptop.
ML Research Handbook connects machine learning foundations with the practical work of reading papers, judging evidence and finding research gaps. It is written in English and Bangla so Bangla-speaking students can meet the ideas in a familiar language.
- Version 4.0: 101 pages, 34 sections and 61 academic references, released as Markdown, PDF and DOCX.
- Covers critical paper analysis, research gap identification, novelty and reproducibility, with an AlexNet case study.
- An educational resource, not original research. I revise it as feedback comes in.
Machine learning · Research methods · Technical writing
Read the handbook → · Why I wrote it
Details of these four projects
Maze Mind · Python · Pygame CE · SQLite · scikit-learn · pandas
- Randomized DFS maze generation and an A* hunter, with classic and isometric Pygame views.
- SQLite records of completed rounds and in-game snapshots for feature engineering and analysis.
- KNN skill classification that drives adaptive difficulty, K-Means playstyle clustering, and logistic-regression escape estimates.
A learning project: bot-simulated training results do not establish accuracy on human players.
Algorithm Atlas · Python · C++ · NumPy · pytest · Actively developing
- 200 algorithms implemented from scratch in Python, 101 of them also in C++, across data structures and algorithms, machine learning and AI search.
- Lessons with dry runs, complexity analysis, input contracts and edge-case tests.
- More than 1,800 pytest tests, run through GitHub Actions.
Machine Learning Hands-On · Python · pandas · scikit-learn · TensorFlow / Keras · Jupyter
- Housing regression with Linear Regression, Ridge and Lasso, evaluated with MAE, MSE and R².
- Iris clustering with K-Means and PCA; data exploration with pandas.
- MNIST digit recognition and a dense-network versus CNN comparison on Fashion-MNIST.
MediSmart · C · ESP32 / ESP-IDF · FreeRTOS · PlatformIO · Firebase
- A university team prototype connecting dual-infusion control and patient monitoring with embedded sensing, actuation and web dashboards.
- MAX30102 and AD8232 sensing, HX711 load-cell input, PWM servo control and a local buzzer.
- Prototype photos · Technical report
Earlier practice: Web development mini projects · SQL & database coursework · All repositories →
The complete list is in my résumé.
Open to academic projects, beginner-friendly web collaborations, and Java or C++ projects.
