Education
BRAC UniversityDhaka, Bangladesh
Bachelor of Science in Computer Science — CGPA 3.76, High DistinctionJan 2020 – May 2024
- Thesis: Towards Myocardial Infarction Diagnosis: An Investigation into Machine Learning and Deep Learning Algorithms using ECG Data.
Supervisor: Rafeed Rahman
Research Experience
Supervised by Annajiat Alim Rasel (NLP and AI security) and by Rafeed Rahman and Nabuat Zaman Nahim (ECG diagnosis), BRAC University
Towards Myocardial Infarction Diagnosis: An Investigation into Machine Learning and Deep Learning Algorithms using ECG Data
Submitted to arXiv
- Proposed Visionary V2, an adaptive ECG filtration method reaching 96.6% accuracy with a Bagging ensemble on all 15 PTBDB leads, and built a custom ResNet-50 on generated ECG images that misclassified 1 of 7,000 validation images.
Subjective Question Generation and Answer Evaluation using NLP
Manuscript submitted for publicationarXiv:2512.17289 ↗
- Developed a comprehensive framework for automated subjective question generation and answer evaluation targeting higher-order thinking skills based on Bloom's Taxonomy.
- Fine-tuned Mistral 7B, Llama-2 7B, and Falcon 7B, demonstrating that Mistral 7B achieves performance comparable to GPT-3.5 in question generation, while GPT-3.5 maintained superior accuracy (ranking top 2 in >96% of cases) for answer evaluation.
Beyond the Benchmark: Innovative Defenses Against Prompt Injection Attacks
Manuscript submitted for publicationarXiv:2512.16307 ↗
- Developed an iterative chain-of-thought defense framework using GPT-4o to automatically generate and refine detection prompts.
- Reduced attack success by 80% and false negatives by 79% on LLaMA compared to baseline.
Detecting Prompt Injection Attacks Against Applications Using Classifiers
arXiv preprintarXiv:2512.12583 ↗
- Implemented and evaluated deep and traditional classifiers (LSTM, FNN, Random Forest, Naive Bayes) for prompt injection detection, improving the robustness of application prompts and downstream systems.