Full Stack Developer · Founder, Stacktern
Building AI products, one shipped idea at a time. Full-stack · ML · Data.
Math, applied AI, and a founder mindset.
I'm a Full Stack Developer with a foundation in mathematics (MSc, SUST), working across Next.js / React on the frontend, Python / FastAPI on the backend, and PyTorch / TensorFlow for ML — shipping real, deployed products rather than notebooks. Alongside engineering, I work in research in the field of machine learning and have completed four papers so far. In March 2026 I founded Stacktern, an independent project hub at the intersection of AI, data, and developer tools, and I'm most useful where the goal is to take something from idea to deployed, working product.
A timeline of roles, internships, and the founder journey.
Building customer-facing platforms end-to-end with Next.js + TypeScript.
Contributing to ML and data science research initiatives.
Independent project hub at the intersection of AI, data, and developer tools.
An independent project hub at the intersection of AI, data, and developer tools.
Founder · March 2026
Building Tomorrow's Stack — AI, data, and developer tooling.
Open-source Python library & CLI that solves math problems step-by-step — linear algebra, root-finding, calculus, and ODEs — verified against NumPy/SciPy/SymPy.
Data pipeline observability with anomaly detection.
Selected work spanning ML, data engineering, and agentic systems.
28 REST endpoints · OpenAPI docs
Full-stack hiring platform: companies build coding, MCQ, and written assessments from a shared question bank, invite candidates to timed attempts, and review auto- and manually-graded results. JWT auth, role-based access (candidate / company / admin), and Stripe-powered paid assessments.
val loss 1.51 EN / 1.66 BN
25M-param transformer in PyTorch, trained on Kaggle T4. Full-stack deployment via FastAPI + Next.js on Hugging Face Spaces.
19-table Delta Lake medallion architecture on Databricks / PySpark. 4-task workflow enforcing dimension-before-fact order.
Multi-agent (Planner / Architect / Coder) system that generates production-ready code from requirements.
R² = 0.9
Gradient Boosting Regression analysis. Distance and peak hours identified as key fare drivers.
Research credibility and continuous learning.
MSc Mathematics
Shahjalal University of Science and Technology
2024 – 2025
BSc Mathematics
Shahjalal University of Science and Technology
2019 – 2024
Parameter-Efficient Fine-Tuning of Language Models for Bangla: A Review of Methods, Models, Resources and Evaluation Practice
A literature review of PEFT methods (LoRA, QLoRA, adapters) for Bangla LLMs, surveying existing models, datasets, and benchmarks, and identifying nine gaps in current evaluation practice.
Mathematical Modeling Specification for Feature-Trustworthiness Selection (FTS): Wearable Stress Detection
Formalizes the math behind a wearable stress-detection pipeline — signal normalization, a feature-trustworthiness score, cross-dataset transfer/confounding metrics, and context-aware heart-rate gating.
Proficient vs. familiar — visual weight matches actual depth.
Open to applied AI, full-stack, and founder-minded collaborations.