Hey,👋
I'm
Harsha R.
I Turn Vision Into Reality With Code & Design
I'm Harsha R., a VIT Chennai Computer Science student passionate about AI and app development. My projects span DeepFake detection to fintech apps, aiming to solve real-world problems with cutting-edge technology
Aspiring AI Research Scientist, I thrive on exploring new tech and tackling cybersecurity challenges to drive innovation and impact.
Hello! I'm Harsha R., a dedicated app developer and AI enthusiast currently pursuing my Bachelor of Technology degree in Computer Science Engineering with Cyber Physical Systems from Vellore Institute of Technology, Chennai. With a solid foundation in computer science, I have honed my skills in app development, deep learning, AI, and NLP.
I am passionate about creating innovative solutions to tackle real-world problems. My projects range from a Speech to Text with AI Correction System to a food waste management app, showcasing my ability to apply technology in meaningful ways. My goal is to become an Artificial Intelligence Research Scientist, driving innovation and making impactful contributions to the tech industry.
Outside of coding, I enjoy exploring new technologies, working on open-source projects, and participating in hackathons. I'm always eager to learn and take on new challenges that push the boundaries of what's possible with technology.
Python
Java
C
C++
HTML
CSS
JS
REACT
Bootstrap
Tensorflow
GIT
React Native
Flutter
I have integrated various essential features for women's safety into apps, ensuring enhanced security and support in everyday life. These features include emergency SOS alerts, safe route planning, and real-time location tracking. My goal is to create technology that provides practical solutions to real-world problems.
A project created through how to train a Transformer model using TensorFlow on the Google - American Sign Language Fingerspelling Recognition dataset made available for this competition(kaggle).
Developed an AI-driven lyrics generator leveraging Hugging Face's GPT model to create song lyrics from user-provided prompts, enhancing creativity and efficiency for songwriters.
Implemented a DeepSpeech 2-inspired ASR model with TensorFlow, utilizing CNNs and Bidirectional GRUs for feature extraction and sequence processing, and CTC loss for sequence labeling. The project covers data preparation, training, and inference.
Developed a model to detect AI-generated text using TensorFlow and Keras, focusing on content from Large Language Models (LLMs). Implemented robust data preprocessing, stratified cross-validation, and result visualization, with detailed experiment tracking via Weights & Biases.
Developed a CNN model using TensorFlow to classify malaria parasites in images. The project includes data preparation, augmentation, model training, and evaluation, leveraging TensorFlow Datasets and optional tools for experiment logging and visualization. The approach emphasizes robust data handling and model performance optimization.
I gained hands-on experience in applying machine learning algorithms, including linear regression and KNN, to real-world problems through projects such as credit fraud detection, movie recommendations, and food calorie counting using computer vision.
During a 2-month internship, I developed a comprehensive hospital management system using JavaScript, HTML, CSS, and PHP. The system streamlined patient and administrative workflows, improving efficiency and user experience. This project enhanced my skills in full-stack development and practical application of web technologies.
In this research, I created a novel DeepFake detection method using CNNs for images and a CNN-RNN model for videos. The models showed improved accuracy, precision, recall, and AUC, enhancing the detection of manipulated media and bolstering digital content trustworthiness.
I compared cybersecurity techniques across multiple parameters, highlighting machine learning's strengths and recommending hybrid approaches.
Conducted research on Parkinson's disease detection using Random Forest, achieving a perfect accuracy of 1.0. Applied Grid Search Cross-Validation to optimize the model's performance and enhance the accuracy. This work contributes to more reliable early diagnosis and effective monitoring of Parkinson's disease.