Akshay Hebbar

Introduction

© 2023 Akshay Hebbar. All rights reserved.

Hello and Welcome to my personal website!

I am thrilled to have you here. My name is Akshay Hebbar, and I am a passionate and driven individual with a background in Computer Science and a focus on cutting-edge technologies. On this platform, I aim to share my journey, experiences, and projects that have shaped my career so far.

Hailing from a humble background, I started my tech journey in my 11th grade when I received my first personal computer in 2011 and yes! it came with a CRT monitor. I have started my programming voyage from TURBO C and have moved all the way up to Java, Python, Frameworks and Machine Learning.

Graduating with a Master's in Computer Science from Syracuse University and a Bachelor's from Visvesvaraya Technological University in Bengaluru, India, I have honed my skills in areas such as Algorithms, REST API's, AI, Data Science, and more. Throughout my academic journey, I have delved into fascinating courses like Data Mining, Reinforcement Learning, and Natural Language Processing, among others and have had the chance to learn from some incredibly passionate people.

Throughout my professional journey, I have embraced challenging projects and actively participated in various innovative endeavors. From crafting and deploying Spring Boot microservices to creating cutting-edge Machine Learning models, I have continuously sought opportunities to make meaningful contributions to the ever-evolving world of technology.

As you navigate through this website, you'll discover a compilation of projects that reflect my love for problem-solving and my eagerness to explore new frontiers. I am always eager to collaborate, engage and learn from like-minded individuals, so feel free to connect with me through my social channels or contact me directly. Let's embark on this technological journey together and make a meaningful impact on the world of Computer Science.

Please do check out some sample of my work here.

Academics

Masters of Science in Computer Science
Syracuse University, College of Engineering and Computer Science
New York, USA August 2021 - May 2023

Courses

• Artificial Intelligence
• Principles of Operating Systems
• Design and Analysis of Algorithms
• Social Media and Data Mining
• Natural Language Processing
• Reinforcement Learning
• Data Science
• Evolutionary Machine Learning

Bachelor of Engineering in Computer Science and Engineering
B.N.M.I.T (Affiliated to Visvesvaraya Technological University), College of Engineering and Computer Science
Bengaluru, India August 2012 - May 2016

Courses

• Algorithms
• Data Structures with C
• Object Oriented Programming with C++
• Discrete Mathematics
• Computer Architecture
• Operations Research
• Computer Graphics
• Computer Networks
• Database Management Systems
• Software Engineering
• Web Programming
• Java Programming

Projects

GitHub Diabetes Onset Prediction using Genetic Algorithm

• In this research, we investigate the possibility of applying a hybrid approach to optimize neural network weights
• Genetic Algorithms are a class of optimization algorithms that perform heuristic search and span out as tree
• Thus, we apply MCTS on genetic tree structure to navigate/guide the heuristic search process of opimizing the neural network weights
• Dataset: Diabetes Classification

GitHub Credit Card Fraud Detection

• Developed a set of Credit Card Fraud detection models and analyzed their relative performance.
• Pre-processed the data with EDA (Exploratory data analysis) and PCA (Principal component analysis)
• Re-balanced the dataset using undersampling (Reservoir sampling) and oversampling (SMOTE) techniques and performed hyperparameter tuning using GridSearchCV
• Applied following machine learning techniques to detect fraud:
  1. Support Vector Machine
  2. Logistic Regression
  3. Decision Tree
  4. Random Forest XGBoost
• Dataset: Credit Card Fraud

GitHub GPT2PDF Download Extension

• Developed a chrome extension using Javascript to download ChatGPT coversations.
• Included an injected download button on the chatGPT platform to download specific conversations as pdf files

GitHub Text to Image GAN

• Developed and trained a Generative Adversarial Network (GAN) to generate flower images from text inputs
• Preprocessed images as numpy arrays and image description into embeddings and appened them to image data
• Trained the Generator and Discriminator networks on the flowers-102 dataset and word embeddings from glove 300D
• Dataset: Oxford Flower 102 and Glove 300D embeddings

GitHub Tweetrimony

• Developed a social media mining application in Python to match users based on their social media actvity
• Generated demographic information using username and tweet history
• Preprocessed data by filtering non user accounts and analyzed tweet sentiments with polarity and subjectivity
• Generated a hash mechanism of multiple secret keys to access data when rate limited
• Datasource: Twitter API, tweets

GitHub ChainReaction AI game

• Developed a two player chain reaction board game in Python with A.I agents as opponents.
• Applied the concepts of adversarial tree search strategies and CNN based approaches in generating the A.I. agent
• AI agent strategies include:
  1. Minimax: Tree based AI
  2. UCT: Uniform Cost search/Random search
  3. CNN with data augmentation

Research

GitHub Monte-Carlo Tree search guided Genetic Algorithm Paper
Hebbar, A. (2023). MCTS guided Genetic Algorithm for optimization of neural network weights. arXiv preprint arXiv:2308.04459, doi: https://doi.org/10.48550/arXiv.2308.04459

Abstract

• In this research, we investigate the possibility of applying a search strategy to genetic algorithms to explore the entire genetic tree structure. Several methods aid in performing tree searches; however, simpler algorithms such as breadth-first, depth-first, and iterative techniques are computation-heavy and often result in a long execution time. Adversarial techniques are often the preferred mechanism when performing a probabilistic search, yielding optimal results more quickly. The problem we are trying to tackle in this paper is the optimization of neural networks using genetic algorithms. Genetic algorithms (GA) form a tree of possible states and provide a mechanism for rewards via the fitness function. Monte Carlo Tree Search (MCTS) has proven to be an effective tree search strategy given states and rewards; therefore, we will combine these approaches to optimally search for the best result generated with genetic algorithms.
• Dataset: Diabetes Classification

GitHub Augmented Intelligence: Enhancing Human Capabilities Paper
A. Hebbar, "Augmented intelligence: Enhancing human capabilities," 2017 Third International Conference on Research in Computational Intelligence and Communication Networks (ICRCICN), Kolkata, India, 2017, pp. 251-254, doi: 10.1109/ICRCICN.2017.8234515.

Abstract

• Recent times have seen an exponential increase in the use of artificial intelligence in numerous regions. Fields like education, transport, finance, and health have made drastic improvements in the last decade; from predicting the stock market prices and driverless cars to predicting cancer cells in human body. Artificial intelligence and Machine learning combined, have shaped the world to be a better place than yesterday. In this paper, I describe a novel approach towards augmenting artificial and human intelligence with the goal of enhancing the capabilities of human activity using adaptive intelligent agents and deep neural networks. Any intelligent system would have come across a situation where human intervention is essential; wherein human intelligence is required for the complete functioning of the agent. This crossover of the worlds is the key to augmenting both human and artificial intelligence. We can enhance the capabilities of both the entities by introducing behavior and context as variables in the cognitive process.

Work Experience

DIEMLife Inc August 2024 - present
Senior AI Engineer

• Working on integrating Mistral-7B LLM for auto template generation using Ollama & Langchain with JSON formatted outputs.
• Engineered GPT prompts with zero and few-shot inference to generate customized user activity tracking templates.
• Implemented and deployed a Microservice on AWS EC2 Docker instance for tracking activities and retrieving user health data from Google Fit API using authentication workflow leading to 10% data enrichment and 5% increase in user signups.
• Configured NGINX reverse proxy service with upstream servers to route and load balance incoming application traffic.
• Containerized application deployment with Docker and constructed automated CI/CD pipeline with GitHub actions and DockerHub resulting in ~20% faster deployment cycles while reducing overall cost by 10% annually.

TruWeather Solutions July 2022 - January 2024
Machine Learning Engineer/Data Scientist

• Performed data cleaning using Python, Pandas, Numpy and Scikit-learn. Analyzed wind component correlations heatmap to develop feature selection strategies, reducing redundancy, improving model accuracy and improving metric interpretability.
• Trained and deployed an LSTM model using Python, TensorFlow & Keras for time series data analysis on AWS. Tuned the model hyperparameters, regularization coefficients, dropouts and applied early stopping to achieve 93% accuracy.
• Performed feature engineering on wind sensor data and exploratory data analysis to detect outliers and apply sampling on gaussian mixture model leading to 25% better wind simulation data per area.
• Built an ETL data-pipeline proof-of-concept using Apache Airflow enabling the collection and generation of wind simulation data reducing the overall TAT by 88%
• Configured a REST API endpoint via AWS Lambda and API Gateway and created a data visualization of the loss metrics
• Architected an Oracle Cloud (OCI) distributed high-performance cluster system (bare-metal HPC Scale-out) to conduct CFD wind flow simulations and distributed ML Training at scale using Spark.
• Contributed to the development of python code for finding the optimal sensor placement locations using entropy formula mentioned in this journal.
• Prepared OpenFOAM simulation data, including velocity, pressure, temperature, and kinetic energy, as input feature vectors for the LSTM model.

DIEMLife Inc (Co-Op) February 2022 - July 2022
Backend Engineer Intern (Data)

• Implemented a fresh SQL schema with graph structure in MySQL to store user health program details.
• Developed a proof-of-concept in Spring Boot microservice on an AWS EC2 docker instance, enabling the retrieval of user health details from Google Fit and mapping activities.
• Architected robust SQL schema, indexing and queries for mapping existing users to their Google Fit healthcare data, enabling data enrichment and enhancing data retrieval efficiency by 10%.
• Triaged platform bugs related to user sign-in, Quest creation and user activity selection to ensure smooth operation.
• Solely managed the end-to-end backend development process at the startup, including system architecture, database design, API development, and deployment, ensuring efficient delivery of critical features.

Envestnet Yodlee July 2016 - May 2021
Senior Software Engineer

• Experience working with cross-functional teams and collaborating across teams for product development.
• [Dream Team Award] - Collaborated with a cross-functional team of Analysts and Software Engineers to design and develop a RESTful microservice architecture using Java, Spring Boot, SQL, JPA/JDBC, and MongoDB (NoSQL) for Account Reconciliation.
• [MVP Award] - Designed a multithreaded SaaS application for account reconciliation, using Kafka and Redis enhancing scalability to handle 60% more reconciliation traffic and achieving 20% improvement in reconciliation rate performance.
• Introduced the concept of synthetic transactions for better account reconciliation process increasing data quality by 24%.
• Developed splunk analytics dashboards and maintained 95% user data quality while helping detect missing account information.
• Implemented test cases using JUnit and Mockito framework, created CI/CD pipelines using perforce streams and Jenkins with SonarQube integration to reduce code smells by 30%.
• Created a time-series MongoDB schema to reduce user metadata footprint leading to 12% reduced storage requirements, enhancing database retrieval speed.
• Led the team in creating transaction type prediction application in python with K-means clustering as part of hackathon.

Contact

Socials

• Twitter
• LinkedIn
• GitHub
• Instagram

Elements

i = 0;

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

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