• Ongoing Research: 

1.  IIT Gandhinagar, Lingo Lab:  Interpretability in Multilingual LLMs

• Previous Research: 
     

1. 1. 1. University of South Carolina: Synthetic dataset generation for Jailbreak prompts using SOTA LLMs.

      2. BITS Pilani: Co-authored a research paper with Prof. Nikumani Choudhury on feature selection in LoRaWAN. 

      3. National Institute of Technology, Trichy:  Adversarial example generation with  FGSM and DeepFool to misclassify finetuned ResNet50.
            
         Details on LinkedIn

• Finalist in multiple national ML and software hackathons, as a team leader.

    ~Amazon ML challenge 2024: 357th rank out of 74,829  [Medium Article]

    ~IIT Dhanbad (HackFest) 2024: Finalist, built AI snipping tool, 35+ GitHub Stars

    ~IIIT Ranchi (House of Hacker) 2023: Pre-Finalist

1. Notable Pull Requests merged: count intersection LoRA finetuning notebook :
    count intersection LoRA finetuning notebook

2. Multiple critical GitHub Issues raised on important repos:

    [Functionality Misssing]: Bringing back the auto detect feature and language feature

    GSM8K score mismatch between offical score and eka-eval

    Claude config.json error

3. Created the following datasets, combined 1000+ downloads on Hugging Face and Kaggle.

• Trilingual lexicon dataset in English, French and Portuguese. https://huggingface.co/datasets/MLap/English-French-Portuguese-Lexicon

• SentiHin-2500 (Hindi Sentimental Analysis dataset).  https://huggingface.co/datasets/MLap/SentiHin-2500

• Book-Scan-OCR https://huggingface.co/datasets/MLap/Book-Scan-OCR

• Sentiment2Emoji https://huggingface.co/datasets/MLap/Sentiment2Emoji

• Evolving Deep Learning Paradigms in Language and Vision (State Space Models, Reasoning Foundational Models)

• AI Safety (Alignment, Adversarial & Jailbreak Robustness, Interpretability)