Kirthana Natarajan

Senior Machine Learning Engineer

• Promoted in August 2024 for impactful contributions to projects and exceptional work ethic. Key contributions include developing and deploying the AI component of multiple real-time computer vision applications and the Yield Optimizer.
• Ideated, developed and deployed the object detection, tracking and segmentation models, OCR and logic layers of multiple video analytics applications, including various workforce SOP compliance monitoring use cases, with 100% violation detection accuracy in live deployment. Optimized the compute usage by over 75% via model and code optimizations.
• Developed a computer vision-based equipment monitoring system to track safety-critical activity of ladles carrying hot molten metal, preventing trolley movement during crane-based ladle loading; successfully deployed at a large steel manufacturer.
• Led the development of the Ripik Optimus Yield Optimizer, which provides optimal process parameter ranges for improving pharmaceutical ingredient yield. Developed the core algorithm based on decision trees. Achieved 3.75\% yield improvement for select molecules across three leading pharmaceutical plants.
• Built the core logic of the Ripik Optimus Manpower Scheduler to allocate manpower based on skill and resource constraints. Deployed across 32 manufacturing blocks.
• Improved glass bottle production efficiency by 25.7% during trial runs at multiple furnaces of a leading glass manufacturer. Developed an approach to identify optimal process parameter ranges, automated recommendation generation, and revamped the data preparation pipeline. Presented to stakeholders to gather requirements and promote adoption.
• Developed a zinc-in-slag prediction model for the Imperial Smelting Furnace at a leading zinc manufacturer, achieving 89% directional prediction accuracy. Built the data engineering backend for the tool.
• Built the dataset downsampling module and custom model training and inference modules of the Ripik AI Studio. Implemented background subtraction methods, hashing-based similar image identification, and ONNX inference.

Undergraduate Researcher

• Worked on a digital solution to detect and relieve stress through multi-modal stress detection using physiological signals such as skin conductance and blood volume pulse, and behavioral data like screen time and phone usage.
• Developed classification models for stress detection by implementing various signal processing techniques, ML and DL model architectures based on research papers. Experimented with normalization techniques, data augmentation, feature extraction and selection, and hyperparameter tuning to improve model performance on lab and field setting data. Achieved 98% accuracy on a public dataset, 70% on cross-dataset validation, 86% in lab settings, and 62% in daily life settings using a custom dataset from seven participants.
• Designed user study protocols for lab and field settings via literature review and conducted data collection studies.
• Developed backend APIs and integrated data streaming from a sensor wrist device to an Android mobile application for data collection.
• Paper submitted to ICASSP 2025 (International Conference on Acoustics, Speech, and Signal Processing).

Software Engineer Intern

• Intern on the Big Data team at Jio Platforms, the digital wing of Reliance.
• Developed data pipelines to move data to Azure Cloud. Used Python, PySpark to automate data quality analysis on Azure Synapse Analytics.
• Learnt about data as a product, data mesh and data engineering concepts.
• Published four comprehensive internal documentation. Received full-time return offer.

Undergradute Researcher

• Worked on developing an early warning system for COVID-19 outbreaks in India on a national scale with district-level granularity, using multimodal data of testing, vaccination, covid strain, clinical and social media data, in collaboration with ICMR (Indian Council of Medical Research).

Software Developer

• Developed backend APIs using FastAPI and MongoDB for mobile applications designed for schizophrenic patients and their caregivers.
• Developed ReactJS components for a web application dashboard displaying the data of patients, caretakers and doctors registered in the system.
• Project in collaboration with AIIMS Delhi. Media

AlgoIllustrated

Interactive data structures and algorithms visualization platform developed using D3.js and Node.js.

SafeNote

Secure Android notes application in Java with Firebase database. Users can create, edit, delete, and protect individual notes. The note content is encrypted and stored securely with cloud backup.

Toysly

Secure e-commerce web application developed using Django, HTML, CSS, and PostgreSQL database. Features include keyword and category-based search, Razorpay payment gateway, OTP-based login/signup, and admin access.

Colour Switch

Spin-off version of the popular game Colour Switch developed using Java and JavaFX. Includes single and multi-player modes, options for sound effects, and a leaderboard.

TwiDash

Fine-grained tweet analysis dashboard created using Flask and JavaScript. Utilizes deep learning models to predict the tone of tweets in real-time from the Twitter API. Provides dispersion networks and visualizations using Plotly.

Transfer Learning on Medical Image Datasets

Explored transfer learning for supervised image classification on medical image datasets BloodMNIST and OCTMNIST from the MEDMNISTv2 database. Performed exploratory data analysis and applied pre-trained models like AlexNet, BEiT, Inception, VGG, ResNet, and SqueezeNet. BEiT model outperformed baseline performance, matching the highest reported accuracy on BloodMNIST.

Human Activity Recognition Using Smartphone Sensors

Implemented and compared various machine learning models, including SVM, logistic regression, random forest, MLP, and KNN, for supervised classification of human activity based on accelerometer and gyroscope data from smartphone sensors, achieving 95.31% accuracy.

Mentor

• Mentored students from economically weaker community.
• Taught English, Mathematics, Science.