Index / Projects
Keywords, highlights and source code for each project.
Fine-tuned an SLM with LoRA to generate synthetic scientific I/O traces.
Fine-tuned SLMs (Granite 3.3B, Gemma) using Hugging Face and LoRA to generate synthetic DLIO HPC I/O traces, enforcing POSIX grammar constraints and evaluating temporal fidelity, scalability, and failure modes for low-cost workload analysis.
Reproduced GPT-2 by pre-training on FineWeb-Edu dataset using PyTorch.
Reproduced GPT-2 using PyTorch and pre-trained on FineWeb-Edu dataset (10B tokens) to surpass Hellaswag performance.
Automated ML training on a streaming/batch telecom churn data using AWS components.
Implemented an automated ML pipeline based on AWS SageMaker, Kinesis, Firehose, and S3 to collect, compile, and pre-process streaming data and train an ML model that detects potential telecom churn with 10K+ data batches.
Architected a distributed file storage system with lxd containerized VMs
Architected a local distributed NFS environment with Ubuntu 22.04 and LXD containerization based VMs to benchmark I/O operations performance for Cassandra and MongoDB. Utilized C, Unix Shell, C++, and Python in the span of the project.
End-to-end NLP pipeline categorizing Covid-19 related tweets.
Implemented an end-to-end NLP pipeline in Python to categorize 6K+ COVID related tweets through crowdsourcing, data preprocessing, feature engineering, and model training with Cohen’s kappa, bag-of-words, and SVM.