Tanmay Mahindrakar

Intro

Hi! I'm Tanmay. I am excited that you are here to know bit more about me. I am a Graduate Student in Computer Software at Indiana University, Bloomington. I am fascinated with the advancements of Software Industry since past couple of decades and the way it is moving forward. I am interested in building large scale software systems with advanced methods to manage enormous amount of data which has direct impact in the lives of society. Feel free to peruse my skills, work experience and awesome projects.

Work

I started my journey as a research intern with Indian Statistical Institute followed by working with the technology giant, Tata Consultancy Services as a software developer. During my pursuit of the Masters program, I had the privilege of collaborating with Ericsson, where I engaged with cutting-edge technology. I undertook a range of responsibilities, which are outlined below.

Data Analyst at Indiana University
• Integrated fine-tuned Large Language Models (LLMs), enhancing natural language processing capabilities for insurance clients using cloud services.
• Synchronized a classification model with Model Pipelining, Exploratory Data Analysis (EDA), Feature Selection, and Feature Engineering using PySpark and Machine Learning, while integrating Tableau dashboards, achieving an accuracy of 93%.
• Optimized complex SQL queries for data retrieval, improving database performance and query execution time by 30%.
• Visualized Key Performance Indicators (KPIs) to identify trends and efficiently extract articles from the knowledge base.

Data Science Intern at Ericsson:
• Developed Large Language Models, including Masked Language Modeling, from scratch. Successfully trained the model on customer telecom data, achieving a strong perplexity score.
• Researched and harnessed fine-tuned GPT-based Causal Language Models and Masked Language Models for downstream tasks, significantly boosting entity recognition accuracy beyond standard Language Models.
• Pioneered a novel approach to Active Learning, resulting in an impressive 30% accuracy improvement and a remarkable 35% reduction in computation requirements. A research paper detailing this method and a patent is currently under review with the Ericsson Research Review Committee.
• Constructed Knowledge Graphs from customer data using Graph Neural Networks with a self-attention mechanism, harnessing this innovative approach for personalized recommendations.

Software Developer at Tata Consultancy Services:
• Developed a shipment tracking and delivery E-commerce software for over 750K+ users using native Android-Kotlin, AWS Dynamo Database, AWS CloudWatch, AWS S3, Postman and Google Firebase.
• Integrated AWS lambda functions using REST API and added Firebase support resulting in 47% performance improvement by providing Statistical Analytics, App Crash Analytics and Remote Configurations.
• Proposed and initiated A/B testing using Remote Configuration with 99.8% crash free experience.
• Mentored junior developer of the team with code reviews and pair programming to get him up to speed.
• Maintained Microsoft Power BI dashboards providing analytical and statistical overview for all stakeholders.
• Collaborated with stakeholders to discuss requirements to improve software design and scalability.

Research Intern at Indian Statistical Institute:
• Implemented a sentimental based analytical project using Clustering which increased precision by 23%.
• Researched and applied Clustering, Classification and Neural Networks algorithms utilizing Sigmoid, Hyperbolic Tangent, SoftMax and Rectified Linear Unit activation functions for 3+ projects.
• Reduced error rate by 30% in an aggressive 3-month deadline for a research-based project for employee attrition prediction.

Projects

My project work covers mostly software engineering as well as data science projects, some of which are noted down below. Do check them out!

Amazon Sentiment Analyser
• Engineered and fine-tuned a custom Sentiment Analysis model for Amazon reviews, achieving over 45,000 downloads on Hugging Face and a vibrant open-source community.
• Deployed and integrated the model into real-world applications, gaining experience in NLP production environments.
• Collaborated closely with both professors and peers to iteratively refine the model, harnessing a diverse range of expertise to elevate its performance and broaden its scope of applications.

Rental Management System
• Led and managed a 7-member team in the end-to-end development of a full-stack web application, delivering efficient rental management system services.
• Orchestrated the integration of advanced features including Google Authentication, Payment Gateway, Google Maps API, and a responsive Chatbot for enhanced user interaction.
• Leveraged Django Python for robust backend functionality and React.js for creating an intuitive and dynamic frontend interface, optimizing user experience with Docker for efficient containerization.

Employee Attrition Prediction.(Technologies: Software Development using Django Framework with Python.)
• Spearheaded an attrition prediction system for robust companies by bypassing the generic mono-pool algorithm selection that achieved flexibility and improved accuracy by 50%.
• Used Support Vector Machine, Gradient Boosting, Naïve Bayes, and Neural Network prediction models.
• Research Paper

Home Credit Default Risk
• Orchestrated and configured classification model with Model Pipelining, Exploratory Data Analysis and Feature Selection using Machine Learning and Deep Learning Algorithm like Multilayer Perceptron, Decision Trees and achieved an overall accuracy of 93% with hyperparameter tuning. Improved base accuracy using SoftMax and Rectified Linear Unit activation function with Tensor Board result visualization.

Path Navigation
• Utilized the concept of breadth first search to find the shortest given a board with obstacles.

Heuristic Search
• Implemented heuristic cost functions and used them to find the shortest distance to the goal state for various problems.

Game Theory and Spam or Ham
• Using game theory, designed an Artificial Intelligence for a game similar to checkers that outputs best moves.
• Applied Naive Bayes Classifier to predict if the given document is spam or not spam.

NLP and Computer Vision
• Utilized the principles of Hidden Markov Models to carry out Optical Character Recognition and Natural Language Processing

Education

Here is my academic journey!

Indiana University, Bloomington
• Coursework: Applied Algorithms, Elements of Artificial Intelligence, Applied Machine Learning, Data Mining, Software Engineering, Database Design, Natural Language Processing
• GPA: 3.94
• Duration: Fall 2022 - Spring 2024

University of Pune
• Coursework: Advanced Data Structures, Programming Fundamentals, Computer Architecture, Operating Systems, Object Oriented Programming, Machine Learning and Artificial Intelligence
• GPA: 8.98/10.0
• Duration: Fall 2016 - Spring 2020

Skills

Here are some of my skills!

Languages
• Python
• Java
• HTML
• C++
• Kotlin

Database
• MySQL
• SQL
• AWS Dynamo DB
• PostgreSQL

Frameworks and Technologies
• Django
• Flask
• Postman
• PyTorch
• TensorFlow
• Google Firecloud/Firebase
• AWS CloudWatch
• AWS S3
• Azure
• Microsoft Power BI
• Git/GitHub
• Bitbucket
• Transformers
• Retrieval Augmented Generation (RAG)
• Large Language Modeling
• Natural Language Processing

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';