Experience

Research Assistant

• Lead author on CLIPByte, a tokenizer-free vision–language model for multilingual open-vocabulary image captioning across six typologically diverse languages (English, Thai, Japanese, Italian, German, Vietnamese).
• Designed cross-modal alignment between a CLIP ViT-B/32 visual encoder and a ByT5-small byte-level decoder, removing tokenizer-induced OOV bottlenecks for low-resource scripts.
• Built the multilingual evaluation pipeline (CIDEr, METEOR, ROUGE-L, BLEU) including a Unicode-safe fix to pycocoevalcap's PTBTokenizer for non-ASCII captions, with NaN-safe averaging and persistent JSON logging.
• Investigating tokenizer-free pretraining objectives, byte-level sequence-length efficiency, and contrastive temperature behavior at character granularity.

Undergraduate Research Assistant

• Led work on multi-teacher knowledge distillation for brain-tumor classification (published, Knowledge-Based Systems, Q1, IF 7.6), targeting resource-constrained deployment with explainable-AI overlays.
• Co-led the Bangladesh 2024 Mass Uprising sentiment study (ICCIT 2025), curating a large-scale multi-platform corpus and benchmarking classical and transformer-based NLP models on context-shifted political discourse.
• Contributed to early-stage exploration of image reconstruction from human brain activity via cross-modal generative models (NeuroCinema-style hierarchical encoders coupled with conditional diffusion).

NSU Chat Assistant — Retrieval-Augmented Question Answering

Built a domain-grounded RAG chatbot for university administrative queries; LLaMA-3-8B generator with Nomic-Embed-Text retrieval over a curated NSU document index.

Transformer-Based E-Commerce Sentiment Analysis

Fine-tuned BERT for three-way sentiment classification on e-commerce reviews; analyzed error modes against classical baselines (SVM, Random Forest, AdaBoost, Gradient Boosting, KNN, Decision Tree).

ML-Based Hybrid Power Production Forecasting

Developed a predictive model for hybrid solar/wind/conventional power generation; benchmarked tree- and ensemble-based regressors for short-horizon forecasting.

Undergraduate Teaching Assistant

• TA for MAT250 (Advanced Calculus), MAT350 (Engineering Mathematics), MAT480 (Partial Differential Equations), and MAT481 (Numerical Methods).
• Conducted 20 hrs/week of office hours; mentored students on conceptual derivations, problem sets, and exam preparation.
• Co-developed course materials with faculty; evaluated student work and proctored 250+ exams across semesters.

Programming Instructor

• Taught Java and C++ from beginner through advanced levels with emphasis on algorithmic problem-solving and DS&A fundamentals.
• Designed lesson plans, hands-on coding labs, and assessment rubrics; provided individualized feedback.

B.Sc. in Computer Science and Engineering

• CGPA: 3.51 / 4.00   (Mark Percentage: 89%)
• Undergraduate Thesis: Decoding Language-Agnostic Image Embeddings.
• Selected Coursework: Machine Learning, Pattern Recognition, Linear Algebra, Probability & Statistics, Numerical Methods, Partial Differential Equations.