Hi, I'm Junaid
🏆 Two "Best Paper" trophies in twelve months. 📊 Three research groups run their analysis on tools I built. 🎙️ Once had 10,000 people tuning in to hear me talk on the radio. Now I make machines talk sense instead.
| 🎯 Doing | Making ML statistically honest — causal inference, fairness, uncertainty, no p-hacking |
| 🏆 Won | 1st Best Paper @ SPECTRA 2026 · Best Paper @ APMEE 2025 |
| 🛠️ Built | StatsPro · ReproHub · a citation-obsessed RAG chatbot |
| 📝 Publishing | 1 journal article, 1 invited book chapter, 7 conference papers, 2 more in review |
| 🎓 Next | Hunting for a PhD in Computational Science / ML / Governance Analytics |
Most ML pipelines quietly skip the statistics part. I don't. I build systems where every claim — a model's accuracy, a policy's effect, a paper's result — has to survive a hypothesis test before I believe it.
That means: causal inference instead of correlation-shrugging, uncertainty quantification instead of point-estimate overconfidence, and reproducibility checks instead of "trust me, it ran on my machine."
Say it in one line: statistically rigorous, causally grounded, reproducible ML — for problems big enough to matter (governance, fairness, public policy).
| Project | Built with | Why it's cool | |
|---|---|---|---|
| ♻️ | ReproHub [live] | Python · SciPy | Re-runs a paper's stats on the raw data and actually checks if the claims hold — not just "is p < 0.05" |
| 📊 | StatsPro [live] | Streamlit · scikit-learn | CSV in, full report out — 10 tests, 6 ML models, AI-narrated insights. 3 research groups run on it |
| 🤖 | Smart RAG Chatbot [live] | LangChain · FAISS | Refuses to answer without citing the exact passage. No hallucinated sources allowed |
| 🤗 | IMDb Transformer | PyTorch · DistilBERT | 87% accuracy, loss dropped 0.44 → 0.16 in 3 epochs |
| 🌍 | Economic Classification | XGBoost | 98.3% accuracy sorting 150+ countries by World Bank data |
| 🔍 | Crime Analytics | XGBoost · SMOTE | 93% accuracy across 6 crime types, spatial-temporal features |
| 🏭 | Enterprise Survey EDA | Pandas · SciPy | ANOVA across 55,620 records, 117 industries (p < 0.001, for real) |
🏆 Award-winning work
- Ahmed, S.J., Islam Nahian, M.T., & Kwoshik, M.H.R. (2026). "CF-EGAT: A Causal Fairness-Aware Equity Graph Attention Network for Country-Level Environmental Livability Classification." Symposium on Photonics, Emerging Computational Technologies, Research & AI-Data Science (SPECTRA 2026). Oral Presentation. 🏆 1st Best Paper Award. DOI: 10.5281/zenodo.21195761
- Ahmed, S.J. (2025). "Multi-Dimensional Statistical Similarity for Governance Classification: Beyond Arbitrary Thresholds in Comparative Politics." 6th Annual Paper Meet Electrical Engineering Division (APMEE 2025). Oral Presentation. 🏆 Best Research Paper Award.
📄 Journal & conference publications
- Ahmed, S.J., Kwoshik, M.H.R., & Islam Nahian, M.T. (2026). "Machine Learning for Crime Classification: A Fairness-Aware Approach to Class Imbalance." Journal of Machine Learning and Applications, 2(1), 9–17. DOI: 10.61577/jmla.2026.100002
- Ahmed, S.J., Islam Nahian, M.T., & Kwoshik, M.H.R. (2026). "RMA-BO: Regret-Minimizing Adaptive Bayesian Optimization." SPECTRA 2026. Oral Presentation. DOI: 10.5281/zenodo.21194394
- Ahmed, S.J., Islam Nahian, M.T., & Kwoshik, M.H.R. (2026). "Environmental Livability Assessment via Adaptive Bootstrap-Retrained SHAP and Statistically-Constrained Pareto Counterfactuals: A Cross-National Analysis." 5th IEEE International Conference on Signal Processing, Information, Communication and Systems (SPICSCON 2026). BAUET, Bangladesh. August 13–14, 2026. Accepted for Presentation. IEEE Xplore.
- Ahmed, S.J., Kwoshik, M.H.R., & Islam Nahian, M.T. (2026). "Machine Learning for Crime Classification: A Fairness-Aware Approach to Class Imbalance." SPICSCON 2026. BAUET, Bangladesh. August 13–14, 2026. Accepted for Presentation. IEEE Xplore. (conference presentation of the journal article above)
- Ahmed, S.J. (2026). "DeepEnMap: Ordinal-Aware Multi-Modal Deep Learning for Energy Poverty Risk Mapping." IEMIS 2026 — 4th International Conference on Emerging Technologies in Data Mining and Information Security. UBC, Vancouver, Canada. August 10–12, 2026. Accepted for Presentation. Springer LNNS Series (Scopus, EI-Compendex, DBLP, ISI Proceedings).
- Ahmed, S.J. (2026). "Density-Decoupled, Mask-Ablated Segmentation-Guided Diffusion for Controllable Mammography Synthesis: A Preliminary Study." IEMIS 2026 — 4th International Conference on Emerging Technologies in Data Mining and Information Security. UBC, Vancouver, Canada. August 10–12, 2026. Accepted for Presentation. Springer LNNS Series (Scopus, EI-Compendex, DBLP, ISI Proceedings).
📖 Book chapter (invited)
- Ahmed, S.J. (2026). "Generative AI and Mathematical Optimization for Football Match Outcome Prediction: A Comparative Study of CatBoost, XGBoost, and TabNet with Kelly Index Stratification." In: Lahby, M. (ed.), Generative AI and Mathematical Optimization for Performance and Innovation in Football, Springer Optimization and Its Applications. Springer, Cham. (Invited Chapter, Under Review)
🔬 Manuscripts under review
- Ahmed, S.J. (2026). "DemocracyGuard: Testing a Divergence-Index Reconciliation of Subjective and Objective Democracy Indicators for Forecasting Adverse Regime Transitions." Under review at Transactions on Machine Learning Research (TMLR). (Q1, Top-Tier Journal)
- Ahmed, S.J. (2026). "FAI: Feature-Wise Adaptive Imputation via Downstream-Aware Method Selection." Under review at ICISET 2026 (IEEE Xplore).
Updates daily · powered by a GitHub Action, not a static image
The math behind it: Hypothesis testing · Bayesian & causal inference · Bootstrap · FDR correction · Uncertainty quantification · Measure-theoretic probability
Researcher, Royal Scientific Publications (2026–present) — chasing open problems in computational science and applied ML.
Senior Researcher, Young Learners' Research Lab, RUET (2022–2024) — ran the lab's hypothesis-testing pipeline, mentored 5+ junior researchers (one shipped their first conference paper because of it), and made "reproducible" the lab's default, not the exception.
On the side: tutored 50+ underprivileged students for free, mentored 4 into RUET/BUET/CUET, guided 5 to perfect board-exam GPAs — because good statistics should help real people, not just impress reviewers.
Before the p-values, there was a radio mic. I hosted a weekly show during COVID that reached ~10,000 listeners, published a Bengali poetry collection ("Sob Odvuture" — 50+ poems, 200+ copies sold), and wrote/performed 15+ original songs. Turns out storytelling and statistical rigor want the same thing: something worth trusting.