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
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
Answers questions about any YouTube video: pulls the transcript, embeds it with Sentence Transformers into a FAISS store, and answers strictly from the retrieved context using Gemini via LangChain.
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