Mashfiq UZ Zaman
Final-year Computer Science and Engineering student
BRAC University
Project
Built a controlled Bangla–English evaluation set of 4,667 articles by prompting an LLM to rewrite real news at three levels of factual change, with automated filtering to validate every generated label. Benchmarked a zero-shot Gemini API judge against fin
Project
Deployed Streamlit app turning a podcast transcript into an episode summary, timestamped chapters and a comprehension quiz, with any single section revisable without regenerating the others. Built and shipped in 48 hours. Gemini 3.5 Flash Lite with an aut
Project
Semantic retrieval pipeline in LangChain, ChromaDB and the Gemini API: a claim returns ranked supporting documents plus a structured JSON verdict with a confidence score.
Project
Fine-tuned XLM-RoBERTa for fake-news detection, reaching 89% accuracy and 0.99 macro-F1 on the evaluation set.