Working on EEG-to-language decoding using large language models to study cognitive signal representations and neural–linguistic alignment.

• Designing deep learning pipelines for EEG preprocessing, time–frequency transformations and multimodal embeddings.
• Implementing and evaluating LLM-based architectures for intent / semantic prediction from biosignals (transformers, contrastive learning, calibration metrics).
• Running experiments on benchmarking EEG datasets; tracking model accuracy, calibration (ECE) and robustness.
• Target outcome: Q1 journal publication under Dr. Yuvaraj Rajamanickam (NTU/NIE) and Dr. Amalin Prince (BITS Goa).

Designing an LLM-assisted system to detect and resolve ambiguous natural language queries in enterprise search spanning databases, documents, and knowledge graphs.

Exploring GPU-Direct Storage (GDS), HDF5 optimisations and high-performance data pipelines for large-scale ML workloads.

• Profiling I/O throughput, PCIe bandwidth and CPU–GPU transfer bottlenecks using NVIDIA Nsight and custom benchmarks.
• Contributing to modifications in h5-bench and HDF5 GDS tooling to support GPU-aware data loading for AI/HPC use cases.
• Building understanding of HPC fundamentals: parallel I/O, memory- vs compute-bound workloads, MPI basics and high-speed storage.
• Long-term aim: integrate these insights into ML training pipelines and feed into future performance-oriented publications.