About
I'm an AI/ML practitioner with a strong interest in research-driven development. I work on designing custom architectures, developing novel approaches to complex problems, and exploring ideas through systematic experimentation. I also contribute to open-source projects and enjoy engaging with emerging directions in machine learning.
Open Source
meta-pytorch/torchtune
Fixed a failure in Torchtune's cosine LR scheduler when training with torch.compile. Compiled optimizers sometimes replace optimizer.step with a plain function that lacks the method attributes that PyTorch LR schedulers rely on. Added a wrapper to make optimizer.step method-like when needed, restoring compatibility for full-finetune recipes using compile: true.
View PRExperience
Intern @ Tech Mahindra
Built an MCP-powered CI/CD assistant for GitHub Actions with an event-driven webhook server, reducing stored workflow data by ~95% while delivering real-time natural-language insights on builds and deployments.
Generative AI Intern @ Zummit Infolabs
Developed an LLM-powered chatbot for pet owners using RAG and Maximal Marginal Relevance, with query parsing for better intent recognition and FastAPI deployment.
Projects
Semantic Segmentation on Cityscapes
Building novel architectures to perform semantic segmentation on the famous Cityscapes Dataset
View on GitHubCloud Cover Detection
Developing multiple novel methods to perform cloud cover detection using a plethora of features
View on GitHubMCP Integration in CI/CD
A pipeline for integrating LLM CI result analysis through the Model Context Protocol (MCP)
View on GitHubGrammar Scoring from Speech Samples
Performing grammar scoring on speech samples using classical ensembles and PEFT-fine-tuned DistilBERT on Whisper transcripts
View on GitHubSentiment Analysis with LoRA
Fine-tuning distilbert-base-uncased to perform sentiment analysis utilizing parameter efficient fine tuning from LoRA
View on GitHub