MS Computer Science & Engineering researcher at the University at Buffalo. Passionate about building inclusive, low-resource language technologies, scalable AI infrastructure, and state-of-the-art multi-task learning models.
I am a researcher and software engineer driven by the desire to build accessible language technologies. Witnessing firsthand how technology can empower individuals facing language barriers, I pivoted from traditional software engineering to computational linguistics and ML.
Currently, under Dr. Nasrin Akhter, my thesis focuses on Multilingual Speech Emotion Recognition using multi-task learning to address data scarcity for low-resource languages. I aim to bridge computational capabilities with linguistic theory to build systems that adapt to users rather than forcing users to adapt to standardized dialects.
Graduate Research Assistant
Software Engineer
Built a RAG-based conversational framework and fine-tuned LLMs on HPC clusters, evaluating 200 synthetic profiles to enhance personalized AAC communication.
Read Paper →Designed an ML pipeline for rhythmic analysis of a 4.1-hour Nattuvangam dataset, achieving 70% classification accuracy on HPC systems.
Read Paper →Leading the end-to-end development of a real-time speech feedback pipeline (Flask) for dialect acquisition. Curating a specialized audio corpus and fine-tuned ASR/TTS models on dialect-specific phonological features using UB's LakeEffect HPC cluster.
Architected a multilingual SER model on NVIDIA H100 nodes using multi-task learning (ASR auxiliary) to boost low-resource classification. Approached SOTA performance via cross-lingual transfer without language-specific fine-tuning.
Built an intelligent search and summarization system over 50,000 Wikipedia articles using a SOLR indexing pipeline on GCP, with zero-shot classification, BlenderBot, and T5 for response generation.
Implemented Q-Learning and Deep Q-Network (DQN) algorithms in the SumoRL simulation environment to prioritize emergency vehicles at intersections, dynamically adjusting traffic phases to cut average wait times by 45%.
Read Case Study →Engineered a distributed sorting algorithm in C++ using MPI for HPC environments, optimizing inter-process communication by eliminating large-scale all-to-all data exchanges.
MS in Computer Science & Engineering (Research Track) • GPA: 3.81/4.0
Courses: Reinforcement Learning, NLP, Information Retrieval, Computational Linguistics, Parallel & Distributed Processing
MS in Computer Application, AI & Data Science | GPA: 8.83/10.0
2020 – 2022Bachelor in Computer Application, Software Engineering | GPA: 7.93/10.0
2017 – 2020