Hello, my name is
I'm a computer engineer who specializes in Deep Learning, particularly in Computer Vision and Robotic Process Automation.
I graduated with a Master's in Computing (Arificial Intelligence Specialization) from the National University of Singapore.
Currently, I am a Data Scientist at Monarch Tractor, working towards the future of intelligent and autonomous tractors in farming. Yes, literal tractors.
Whether you have a question or just want to say hi, my inbox is always open!
Hello! I'm Saisha.
I enjoy building things that have a real-world purpose. My interests lie at the intersection of Computer Vision and Robotics, all of which began in 2018 when I worked on my first ever project in the field.
As a curious computer engineer, ever fascinated by the puzzles in the world around me, I strive for challenging opportunities to exercise and build upon my repertoire of computation, development, and collaboration skills, and aspire to combat problems that result in positive impacts on the community.
I recently graduated with a Master's in Computing (Artificial Intelligence Specialization) from the National University of Singapore.
Currently, I am a Data Scientist at Monarch Tractor, working towards the future of intelligent and autonomous tractors in farming.
Some tech I work with daily:
Spearheading development of deep neural networks to drive electric autonomous tractors, facilitating precision agriculture and operational efficiency around the world.
Developing a UNet model, fine-tuned on a custom built dataset with image augmentation and stable diffusion for camera clarity detection to enhance operation of complimentary AI models.
Engineered ClearNet, a custom UNet-based model with dual segmentation (87% Mean IoU) and classification (98.6% accuracy) heads for camera clarity detection, crucial for reliable operation of complementary AI models in varying field conditions.
Deriving an advanced semantic segmentation model based on Mask2Former, improving perception mapping accuracy by 35% and enabling auto-labeling, thereby reducing manual annotation time by 60%.
Testing and deploying an object detection model based on YOLOv7 to encompass edge case object detection.
Collaborated with a team of developers to build the proprietary software – a web enabled application with a ReactJS (with Typescript) frontend integrated with a .NET Core backend. Interfacing via a web API, this application manages all the data and functions for a maritime company.
I specialized in building intelligent software components to enhance the daily workflow in maritime companies.
Enhanced user experience and accessiblity with a customizable dashboard written in ReactJS and Typescript. The dashboard allows users to build different types of charts with customizable parameters and drill down options. The dashboard recommends useful charts to users based on past activity on the platform.
Spearheaded research and implementation of state-of-the-art solutions for hateful meme detection, a challenging classification problem.
Designed and developed a robust pipeline to construct a multilingual dataset of images using Python and OpenCV for efficient classification.
Led research and planning efforts for misinformation detection and countermeasures in multilingual and multimodal settings, contributing to the advancement of cutting-edge techniques.
Conducted comprehensive surveys and analyzed state-of-the-art literature and datasets in the domains, collaborating with a team to formulate and refine problem statements for a competition. Successfully prepared and curated high-quality datasets.
Build end-to-end systems using ML and statistics along with commercial tools in Python like Tensorflow, PyTorch, Caffé etc.
Research and development to build a robust knowledge graph extractor from unstructured text in compliance documents to assess client's asset justification. Tested several methods including customization of proven models like BERT, linguistic analysis, etc.
Built a recommendation system along with two other colleagues in Python, with a collaborative, content-based and hybrid model. Gathered customers feedback to train and further improve the system and add additional features to improve user experience.
BAU involved labeling and analyzing large sets of visual (images and videos), gathering customer feedback and building new models and improve, training ML models and writing code to prepare, use, enhance and test models, and researching public ML models and datasets.
Research and planning for the misinformation detection and counter problem in multilingual and multimodal settings.
Surveyed and reported state-of-the-art literature and datasets in the domain.
Collaborated with a team to formulate and refine a problem statement for a competition in the domain and prepared a dataset for the same.
Worked on a project, Quantity AI, which leverages computer vision techniques to extract information from CAD drawings and plans
Debugged and fixed model errors, as well as deployment issues in the pipeline for the Quantity AI web application
Communicated and collaborated with multi-disciplinary teams of engineers and designers daily to build a sustainable pipeline to generate 2D drawings from 3D models.
Developed a 3D CNN for part recognition from voxelized 3D CAD models (92% accuracy). Implemented a transfer learning approach for part, shape, and bracket recognition using 2D images taken from different orientations of 3D models (99% accuracy).
Formulated a robust API design to interface with deep learning solutions hosted on AWS, composed microservices using Docker, automated image capture of 3D models and encoding images into JSON, amongst other tasks to establish a smooth process flow, and reduced required pipeline time by 20%.
Expedited development time (20%) by building a robust and load-balanced integration of iOS and Android mobile applications with external ML APIs hosted on Azure cloud.
Established minimum viable product for Kratos Android Application and its interface with external APIs with Java and Android Studio.
Built end-to-end APIs with SpringBoot and MongoDB, integrated with the iOS application.
Architected and developed scalable deep learning architectures including autoencoder neural networks and anomaly detection models amongst others, to detect and predict outliers in machine parameters during functioning with 96% accuracy.
Built a deep learning solution to predict when tools attached to a machine are wearing (98% accuracy).
Constructed a microservice tool to aid development by enforcing a constant template for login, authentication, and other common operations in cross-organization applications. The tool works as a Visual Studio Code extension written with Node.js, TypeScript, and Python, and optimizes development time by 20%.
Investigated and revamped automated testing tools for various products- increased speed and reliability by 40%.
Assisted in development, testing and deployment the Android application "TraveLibro" – a platform for travel bloggers.
Conducted trial runs to ensure instruction correctness and production of desired information.
Storage, retrieval and manipulation of data for analysis of system capabilities and requirements.
Master's in Computing (Artificial Intelligence specialization)
GPA: 4.35/5
Courses: AI Planning and Decision Making, Neural Networks and Deep Learning (1 & 2), Uncertainty Modeling in AI, Big Data Analytics, Information Visualization, Pervasive Computing Technology and Solutions, Interaction Design with Virtual & Augmented Reality, Human-Centred Intelligent Systems, Knowledge Discovery & Data Mining
Bachelor of Engineering in Computer Engineering
GPA: 9.39/10
Academic wards: Principal’s Excellence, Academic Excellence, Computer Engineering Branch Ambassador
Extracurriculars: Soccer Captain (2016 - 2019), Entrepreneurship Development and Professional Development Director @ The Rotaract Club of Thadomal Shahani Engineering College (2018 - 2019)
Indian School Certificate in Science (ISC Science)
Grade: 95.56%
Extracurricular Activites: Badminton Captain, Soccer, Cricket
Here are my personal and coursework projects!
In my free time, I love working on projects to exercise and build upon my skills.
A research project on the design and evaluation of live captioning augmented reality (AR) glasses for individuals with hearing impairment.
[Prototype] A smart fridge using IoT component and ML which detects items present inside the fridge. With a Raspberry Pi and accessories, the IoT component collects and relays data to the cloud. An Android application retrieves and presents this data and detected objects to the user.