Hi, I'm a Data Scientist who builds ML models and ships what works.
Data Science graduate student in Berlin. German and Bangladeshi citizen. No visa sponsorship needed. I build with Python, PyTorch, and deep learning.
About Me
I'm Shahabub Alam. Friends call me Nabid. I'm finishing an M.Sc. in Data Science at the University of Potsdam, after coursework at TU Dortmund. I live in Berlin and work on machine learning, data science, and AI. I publish papers and ship projects.
- Berlin
- German citizen (no visa sponsorship)
- EN · DE B2.1 (progressing to B2.2) · Bangla
- Open to DS / ML / AI
Recently I have worked as a Research Assistant at DFKI and as a Student Assistant at TU Berlin and ESCP Business School.
If that sounds useful, say hello in the contact section.
Work Experience
Recent roles first. Older roles are under See more.
Research Assistant
Student Assistant
Student Assistant
Student Assistant
Education
University degrees first. School certificates are listed too.
M.Sc. Data Science
Completed coursework at TU Dortmund University (Oct 2020 - Sept 2023). Same master’s path, continued at Potsdam.
B.Sc. Computer Science and Engineering
Higher Secondary Certificate (Science)
Secondary School Certificate (Science)
My Projects
Selected work with a short case study for each project.
Robust Bengali Speech Recognition Under Domain Shift
Bengali ASR on OOD-Speech (~1,178 hours, 22,645 speakers). Frozen regional Whisper reaches mean WER 0.288 versus Whisper-small at 1.177 (about 76% relative drop). GroupDRO fine-tuning reaches 0.174 (about 40% further relative drop vs frozen). Cloudflare transcript gallery on the case study. Full Gradio/FastAPI weights on request.
LLM From Scratch
Tiny character-level decoder-only Transformer written without nn.Transformer. Causal attention, next-character prediction, and an LSTM baseline on the same public-domain Shakespeare excerpt. Teaching UI and full source on request. Not a production LLM claim.
Legal/Medical Research Engine
Search over 65,000+ legal and medical documents with cited answers and confidence scores. Queries are not stored. Covers EU legal documents and medical guidelines. Try a live query on the case study. Full index on request.
Demand Forecasting System
Supply-chain demand forecasting with LightGBM, XGBoost, classical baselines, and an optional agentic model picker. On a fixed synthetic seed, one-step XGBoost reaches about 26% MAPE, slightly ahead of a naive-mean baseline. Cloudflare forecast demo on the case study. Full tree stack on request.
German Energy Shift
Hybrid shift recommender for DE-LU day-ahead prices (SMARD) and When2Heat heat-pump stress. Published 2022 holdout ridge MAE about 16.91 EUR/MWh. Cloudflare scenario demo on the case study. Full training stack on request.
Marketplace Discovery Platform
Session feed ranking and hybrid search on public marketplace event logs. Feed v1 reaches NDCG@10 about 0.42 versus popularity at 0.12 on the sample holdout. Search v1 reaches NDCG@10 1.00 on curated queries. Frozen Feed and Search demo on the case study. Full FastAPI stack on request.
Research Publications
Papers and conference work in machine learning, computer vision, and NLP.
2025
Physics-Guided Deep Learning for Heat Pump Stress Detection: A Comprehensive Analysis on When2Heat Dataset
2025 International Conference on Decision Aid Sciences and Applications (DASA)
View publicationReal-Time Traffic Sign Detection for Autonomous Vehicles Using YOLOv11
2025 International Conference on Electrical, Computer and Communication Engineering
View publication
2024
Hotel Booking Cancellation Prediction Using Applied Bayesian Models
2024 International Conference on Decision Aid Sciences and Applications
View publicationStatistical and Machine Learning Analysis of Maternal Smoking's Impact on Neonatal Birth Weight
2024 International Conference on Decision Aid Sciences and Applications
View publication
Skills
Tools I use for ML, data work, and projects.
Core: Python · PyTorch / TensorFlow · SQL · ML / DL
Hover a skill to highlight related projects.
Programming & Scripting
Machine Learning
Data Science
Data Analysis
Cloud & Data Engineering
Full-Stack Development
Community (Side Project)
A personal hobby alongside my studies. Not my career focus.
In my free time I run Nabid In Motion, where I teach AI and Machine Learning fundamentals mostly in Bangla, with English captions so a wider audience can follow along. I also publish English videos on tech topics from time to time. Teaching forces me to simplify models, debug live, and write clearer explanations. The goal is simple. Learn deeply, teach clearly, and encourage others to teach next, so the community keeps growing.
To support that, I built a free Study Hub as a static site with an open ML curriculum. Progress stays in your browser with local storage, so no account is needed. You can read lessons and build projects right away. This work helps me practice explaining technical ideas and shipping useful tools. It does not replace my priority of joining a full-time engineering team.
- 21k+ learners on YouTube
- Bangla + English captions
- Free Study Hub
- Learn, then teach
Get In Touch
Email me or message me on LinkedIn. I read both.
Prefer email? Write to contact@shahabub.com.