My journey

  • Actively seeking full-time Software Engineering/Machine Learning opportunities to drive innovation and create impactful solutions.

    Available to join immediately

  • Multimodal LLM for Assistive Technologies
    Enhancing Real-Time Interaction for the Visually Impaired

    Developed a Multimodal LLM to enhance assistive technologies for the visually impaired, incorporating federated learning for privacy and secure model training.

    Implemented federated learning for enhanced data privacy, achieving scalable training within 45 hours on two NVIDIA A100 GPUs, and utilized 4-bit quantization for efficient inference on 8GB GPUs.

    Integrated Speech-to-Text and Text-to-Speech capabilities to facilitate seamless interaction, enhancing the model's usability for visually impaired users.

    Multimodal LLM
  • Floppy
    Federated Learning on a Peer to Peer system

    Developed Floppy, an algorithm leveraging asynchronous Federated Learning integrated with Secure Multi-Party Computation (SMPC) to decentralize AI training and preserve data privacy.

    Ensured robust and efficient learning while safeguarding sensitive data, enabling scalable and secure AI solutions across diverse devices.

    Technologies
  • Enigma
    Awarded "Best Blockchain Project" by Foundry

    A blockchain-based distributed application addressing data privacy concerns in dataset collection.

    • Leveraged Hyperledger Indy to implement a decentralized identifier (DID) allowing users to retain control over their data.
    • Implemented smart contracts for verifying and validating datapoints during user interactions, ensuring privacy while creating reliable datasets.
    • Introduced a rewarding mechanism to incentivize data contributors, fostering a collaborative and privacy-preserving approach to dataset acquisition.
    More details
  • DIGI-BOT

    AI powered chatbot application designed to help customers to help them navigate to relevant solutions for their queries.

    • Spearheaded the development and deployment of the 'DIGI-BOT' chatbot project for 48 locales, leveraging Angular, ReactJS, and Java Spring Boot, handling 300K users per month with a contact avoidance rate of 42.3%.
    • Enhanched customer experience using Natural Language Processing (NLP): Explored and integrated AWS Kendra to the project to add intelligence capabilitites to chatbot. Later on built in-house Question Answering capabilities using transformer-based models such as BERT model leveraging Deepset's Haystack framework.
    • Enhanced user experience by adding Live chat with agent fucntionalities with the help of Salesforce integration
    • Set-up containerization, and handled CI/CD for the project using AWS services such as EC2, Lambda, EKS, API Gateway, and MongoDB
    • Recipient of β€œAward of Excellence” in Quality Award 2021 and β€œBest Performer of the Quarter” in 2022
    Sony Support site
  • MaaS : Mobility as a Service

    A blockchain based solution for smart city plan

    • A challenge program proposed by the Netherlands Ministry of Infrastructure and Water Management.
    • Developed a high-speed processing blockchain common DB for MaaS, a next-generation mobility service that integrates multiple transportation entities.
    • Implemented high-speed data processing capability system using AWS Kinesis Data Stream and Amazon RDS to record and share anonymized travel history and revenue allocation for more than 7 million users per day, using blockchain technology for secure, transparent and reliable data management.
    • Developed an Angular dashboard integrated with AWS Cognito for secure user authentication and Quicksight for real-time data visualization and analytics, fostering smart city plans by providing insights on movement records and revenue allocation for data processing.
    Sony Press release
  • Enigma

    Crafted a tool for a in-house competetion called Code Premier League

    • A tool capable of measuring the performance and productivity of employees which also safeguards their privacy with the help of blockchain technology. .
    • Built desktop application that could clock in employee attendance, track project management tickets, give ingihts to improve their productivity and a anoynmous general insights to employer on overall performance
    • Hyperledger Fabric framework is used to develop a blockchain network to store anaoymous private data of employee
    • Top #5 finalist at Sony
  • Music Genre Recognition

    Predicts genre of the music powered by Machine Learning

    • Developed this project using Deep Neural Network with Keras API.
    • This tool is capable of predicting the genre of any music by analyzing its spectrogram and then training the Keras model built using these spectrograms.
    • Flask web framework is used to create an endpoint to this model.
    More details
  • LRCaps

    Research intern, Indian Institute of Technology (IIT), Mandi

    • Explored the problem of activity recognition from the videos using a Long-Term Recurrent Convolutional Network (LRCN) and Capsule Network (CapsNet).
    • Proposed a new model called Long-Term Recurrent Convolutional Network and Capsule Network for activity recognition (LRCaps) which achieved 94.7% accuracy.
  • Y-Link

    a Youtube Video Downloader

    • A python course project where I developed a Django application which downloads youtube videos when a user provides youtube link.
    More details
    Technologies
  • EduCart 2.0

    An android app for students

    • An android app called EduCart 2.0 which is a portal for students to prepare for technical interviews from course notes and question banks provided in the app and take up mock tests.
    • Total number of downloads on Google Play Store is 2000.
    More details
    Technologies
  • Blockchain Hackathon

    To tackle a challenge called Real-time record maintenance using Blockchain.

    • Participated in 48 hour Hackathon team event organized by Govt of Karnataka, where we developed the solution to challenge a problem called Real-time record maintenance using Blockchain.
    More details