About
My Introduction
I'm an AI enthusiast with 3 years of experience in Machine Learning and Deep Learning, and 1 year of experience in Generative AI. Currently, I work as a Data Scientist at Deep Neura Technologies, where I focus on building a chatbot using Large Language Models. I'm also actively exploring and interested in fine-tuning these models to enhance their performance for various applications. My passion for AI drives me to explore new technologies, solve complex problems, and create innovative solutions. If you're looking to collaborate or need assistance with AI projects, feel free to reach out!
Skills
My technical & miscellaneous skillsAI & Data Science
3+ Years XPMachine Learning
Natural Language Processing
RAG
Deep Learning
Fine-tuning
LLMs
AI-Agents
Programming
3+ Years XPPython
SQL
Frameworks
2+ Years XPPytorch
LangChain
vLLM
Database
2+ Years XPMySQL
MongoDB
Misc
2+ Years XPGit
Postman
CI/CD
AWS
Flask
Docker
LLM-Inference
Experience
My journey in the academic & professional frontB.E - Computer Science
K. J. Somaiya Institute of Technology, Mumbai, IndiaClass XII
Maths, Physics, Chemistry, Biology | K.J somaiya college of science and commerceClass X
N.E.S High School, MumbaiData Scientist
Deep Neura TechnologiesJr Data Scientist
Care Risk SolutionServices
Services that I offerArtificial Intelligence
Projects
My works, projects & contributions
Machine translation
This project showcases the creation of a machine translation model that translates English to Hindi, leveraging Transformer architecture. It explores essential components, including multi-head attention, cross-attention, masked attention, positional embedding, and feed-forward neural networks. The article provides a comprehensive implementation in PyTorch along with optimization strategies to improve model performance.
View
Fine-Tuning LLM
We fine-tuned the Meta Llama-3.1–8B model to simplify Indian legal texts, including BNS sections and public & administrative laws.Using LoRA with Unsloth, we optimized training on e2e Networks' GPUs. To improve clarity, legal texts were rewritten in plain English with Gemini 2.0 Flash and structured in Alpaca-style instruction-response pairs.The model was deployed via vLLM, providing fast, scalable access through a REST API, making legal information more accessible to lawyers, students, and citizens.
View
BollywoodLyricsGPT
Developed a custom GPT-2 model utilizing Karpathy's implementation, specifically tailored for generating coherent Hindi lyrics. The model was trained on a dataset of 10,000 Bollywood lyrics, featuring 124 million parameters and 300,000 tokens. It incorporates 12 decoder layers for enhanced language understanding and employs the GPT-2 tokenizer, showcasing advanced NLP and deep learning techniques in PyTorch.
View
Cricket Analysis of IPL Data
This project involves a comprehensive analysis of IPL data, focusing on various aspects of batsmen and bowlers across different match situations. By employing diverse visualization techniques, I aim to uncover valuable insights into player performance, helping to understand trends and patterns that influence game outcomes.
ViewBlog
My Blogs
Machine Translation using Transformers
This article details the development of an English-to-Hindi machine translation model using Transformer architecture. It covers key components such as multi-head attention, cross-attention, mask attention, positional embedding, and feed-forward neural networks, providing a full PyTorch implementation and optimization tips for enhancing model performance.
View
Simplifying Indian Law with Fine-Tuned LLMs
We fine-tuned the Meta Llama-3.1–8B model to simplify and interpret Indian legal texts, covering Bharatiya Nyaya Sanhita (BNS) sections and public & administrative laws. Using LoRA with Unsloth, we efficiently trained the model on e2e Networks' GPU infrastructure. To enhance accessibility, we rewrote complex legal texts in plain English using Gemini 2.0 Flash and formatted data in Alpaca-style instruction-response pairs. The fine-tuned model was deployed via vLLM for efficient inference, enabling users to query legal information via a scalable REST API. This AI-powered system makes Indian legal knowledge more accessible to lawyers, students, and citizens alike.
View
AI Code Assistant
Created a VS Code extension for efficient Python code generation by fine-tuning the 4-bit quantized Phi-3 model using LoRA with the Unsloth library on the iamtarun/python_code_instructions_18k_alpaca dataset. Optimized the SLM for fast and accurate local execution.
View