About Me
Hi, I’m Sujith, a Master’s student in Applied Computer Science at Concordia University with a strong foundation in machine learning, NLP, and full-stack development. I enjoy building AI-powered solutions that move beyond experiments and deliver real impact in production.
During my academic journey, I’ve balanced research and practical engineering:
🔍 Research: As a research intern at the University of Technology Sydney, I co-authored a multilingual argumentation analysis paper, analyzing 3,400+ social media posts in English and Tamil using transformer-based models (mBERT, XLM-RoBERTa). I automated data pipelines, designed ensemble voting strategies, and performed deep EDA with Pandas, NumPy, and Seaborn.
🎙️ Applied AI: I fine-tuned and deployed OpenAI’s Whisper for speech-to-text in low-resource languages, building an end-to-end inference pipeline with FastAPI, a React frontend, and GCP for scalable cloud serving.
💻 Software Development: I independently built a secure finance system with React + TypeScript, an ASP.NET backend, JWT auth, and CI/CD with Docker and Jenkins. I also developed TubeLytics, an analytics tool for YouTube channels using Play Framework (Java), asynchronous APIs, and custom data visualizations.
I’m passionate about bridging AI research and software engineering, and I love deploying real solutions that solve problems for diverse users. I’m always learning new frameworks, experimenting with novel architectures, and contributing to systems that combine NLP, deep learning, and modern web tech.
I’m actively seeking Fall 2025 co-op or internship opportunities in AI/ML, NLP, or full-stack development, where I can bring my research mindset, coding skills, and practical deployment experience to help teams build impactful products. 📫 Let’s connect! (social) I’m open to collaboration, mentorship, and new opportunities to push the boundaries of what AI and software can do together.
Skill Set
Programming Languages
Python, Java, C, C++, C#, JavaScript, TypeScript, HTML, CSS, PHP, R, Erlang
Web & Frameworks
React, Node.js, Express.js, Angular, .NET, ASP.NET Core, MVC, Spring Boot, Spring MVC, Scala, Flask, FastAPI
Machine Learning & Data
PyTorch, TensorFlow, Keras, NLTK, Numpy, Pandas, OpenCV, Matplotlib, scikit-learn, Tableau, Hugging Face, Google Colab, Jupyter Notebook, Kaggle, MLFlow
Tools & Platforms
.NET Framework, Docker, Kubernetes, Jenkins, Git, GitHub, GitLab, JIRA, Visual Studio, VS Code, IntelliJ, Selenium, VMWare, Linux, REST API, OAuth, JWT
Databases & Cloud
SQL, NoSQL, MySQL, PostgreSQL, MongoDB, GraphQL, Firebase, Azure, GCP, AWS
Practices & Expertise
Agile (Scrum), CI/CD, Object-Oriented Design (OOD), Data Structures, Design Patterns, Unit Testing, Software Testing, Natural Language Processing, Computer Vision, Machine Learning, Deep Learning, Artificial Intelligence, Generative AI, MLOps, LLMs, Fine-Tuning
LeetCode
163 problems solved (102 medium and hard problems)
Work Experience
Research Intern (Machine Learning)
(Aug 2023 – Jan 2024)
Argumentation Analysis On Twitter and YouTube Comments Using Ensemble Learning Of Transformers
University of Technology Sydney
- Co-authored a research paper on multilingual argumentation analysis, analyzing 3,480+ tweets and YouTube comments in English, Tamil, and code-mixed text for nine discourse attributes like stance, relevancy, and tone.
- Developed and fine-tuned transformer-based NLP models (mBERT, XLM RoBERTa Large, XLM-MLM-100-1280) for multilingual text classification using pyTorch and huggingface, achieving up to 12% higher F1 score than traditional SVM and logistic regression.
- Performed Exploratory Data Analysis (EDA) using NumPy, pandas, matplotlib, and seaborn to uncover key sentiment trends and bias patterns, improving annotation guidelines by 15% agreement in later rounds.
- Designed an ensemble voting mechanism combining majority and single-class voting to boost classification precision, improving misclassification handling for code-mixed data by 10–15%.
Projects
AI & ML Projects
ASR for Indian Languages Github Repo →
I developed an advanced automatic speech recognition (ASR) pipeline tailored for four low-resource Indian languages by fine-tuning OpenAI’s Whisper-medium model. To overcome high word error rates typical for dialectal variations and noisy audio, I implemented an ensemble correction module combining a masked language model (XLM-RoBERTa) for context-aware predictions and fine-grained word-level alignment using Difflib. The result reduced the average WER from 42% to 35%. The complete system included robust REST APIs built with FastAPI and Flask, containerized with Docker, and deployed on AWS (S3, Cloudfront and ECS) following MLOps best practices.
PyTorch MLM Whisper AWS Docker React FastAPI Flask Git
Marine Species Detection Try Colab Notebook →
For a challenging underwater marine species detection problem, I built a custom semantic segmentation pipeline from scratch in PyTorch using a UNet architecture augmented with a novel ConvMixer block for improved multi-scale feature extraction. This enhancement boosted the mean intersection-over-union (MIOU) from 78% to 89%. The entire workflow included custom dataset preprocessing, model training on Google Cloud, and inference APIs served via both FastAPI and Flask. I containerized the solution with Docker for scalable deployment and integration with external applications.
PyTorch UNET GCP Docker React FastAPI Flask Git
Software Projects
Finance Portfolio System Github Repo →
I designed and developed a responsive full-stack finance application that helps users track real-time stock market trends and manage personalized investment portfolios. The frontend was built with React, TypeScript, and Tailwind CSS for clean, device-friendly interfaces. On the backend, I used ASP.NET Core to create secure RESTful APIs with JWT-based authentication and real-time stock data integration through external financial APIs with Axios. The system’s data integrity was ensured using well-structured SQL databases. For scalability and maintainability, I containerized the application using Docker and deploymed the app on azure. Azure SQL was used as a database and Azure cache for redis service for user session caching.
React TypeScript TailwindCSS ASP.NET SQL JWT Docker Redis Azure
Youtube Lytics Github Repo →
TubeLytics is a full-stack analytics platform I built using the Java Play Framework to provide deep insights into YouTube video content. By integrating the YouTube API and designing a reactive server-push architecture with WebSockets and Apache Pekko actors, I enabled live keyword searches, real-time trend tracking, and instant updates for concurrent users. Core features include dynamic channel profiling, word and tag frequency analysis, sentiment scoring, and robust session handling. Functional programming principles with Java Streams and comprehensive unit testing with JUnit and Mockito ensured the system’s efficiency and reliability.
Play Framework Java Websockets SBT Maven Git JUnit Mockito Javadoc
Ptorts - Pet Adoption Platform Github Repo →
Ptorts is a social media-driven pet adoption platform I created using the MERN stack (MongoDB, Express.js, React, Node.js). It allows pet owners to upload videos and images of pets available for adoption, leveraging Firebase Storage for scalable media hosting. To support direct community contributions, I integrated secure crypto wallet functionality with Polygon (MATIC), enabling transparent donations directly to pet owners without intermediaries. The app’s backend APIs manage user authentication, video streaming, and wallet transactions, while the responsive React frontend and Redux state management deliver a smooth user experience for browsing, liking, and donating.
React Node.JS Express.JS MongoDB Firebase Redux React Routers
Education
Master of Applied Computer Science
Concordia University, Montreal (Sept 2024 – Aug 2026)
Bachelor of Technology in Computer Science
Vellore Institute of Technology, Chennai (Sept 2020 – May 2024)
