Kayne Maniti
Software & Machine Learning Engineer
Software Development
Machine Learning
High Performance Computing
Whats up?
About Me
Whats up! I am a passionate Machine Learning and Software Engineer with hands-on experience in AI research, neural network development, and software engineering..
- •Specializing in AI research, deep learning, computer vision, and natural language processing (NLP)
- •Familiar with working in team environments using SCRUM for agile project management and CRISP-DM for data-driven decision-making.
- •A 2nd year Computer Science student at the University of California San Diego
- •When not coding: I like to read manga, drink coffee, build keyboards (i know classic), and I also have a strong passion for teaching, so I may be posting tutorials soon.
Experiences
Machine Learning Engineer Intern
DirecTV
Aug. 2024 - Dec. 2024
Led technical portions of the team, focusing on NLP and sentiment analysis for business insights.
Key Achievements:
- •Led a team of developers to evaluate customer sentiment in Yelp reviews using NLP
- •Parallelized sentiment analysis workflows to optimize performance by 40%
- •Implemented large-scale sentiment analysis paired with regression models for business insights
Software Developer Intern
Health Futures Lab
Jan. 2025 - Present
Conducting research on remote photoplethysmography (rPPG) technology and valence emotion detection to enhance augmented reality-assisted smart healthcare systems.
Key Achievements:
- •Developed rPPG-Playground, a Python and OpenCV-based platform hosting pre-trained PPG models like PhysNet for remote photoplethysmography signal processing
- •Optimized valence emotion detection models, improving real-time classification accuracy and system efficiency
- •Integrated Google Cloud APIs to enhance data processing and scalability for emotion recognition models
Machine Learning Researcher
UCSD Computer Science & Engineering
Oct. 2023 - Dec. 2024
Assisted in the development of machine learning models, drone technology, and DevOps, contributing to the research on avocado tree phenotypic features.
Key Achievements:
- •Collaborated in an Agile SCRUM environment, deploying OpenProject with HTTPS for secure project management
- •Assembled drone cameras and integrated NVIDIA Jetson Nano for hosting quantized AI models
- •Developed software for self-piloting drone systems to collect phenotypic features from avocado trees, implementing MavLink communication protocols
- •Trained and tested DNN models like YOLO, SegNet, and DeepLab for object detection and segmentation using TensorFlow and PyTorch
Artificial Intelligence Fellowship
Break Through Tech @ UCLA
May. 2023 - April 2025
Selected from 3000+ applicants for an intensive AI fellowship program, applying CRISP-DM methodology for data-driven projects, with hands-on experience in model training, data preparation, and analysis.
Key Achievements:
- •Trained CNNs and sentiment analysis models using Keras and TensorFlow.
- •Utilized the CRISP-DM methodology to guide model development and data analysis, ensuring actionable insights.
- •Participated in a Kaggle competition for skin lesion detection, collaborating with a team to develop accurate models for medical image analysis.
- •Received a $2000 stipend for successful completion of the fellowship program.
Instructional Assistant / Tutor
UCSD Computer Science & Engineering
Aug. 2024 - Sep. 2024
Tutored over 100 students in Python programming, helping them grasp key concepts in algorithmic thinking, data structures, and problem-solving.
Key Achievements:
- •Tutored 100+ students in Python programming, focusing on algorithmic thinking and problem-solving.
- •Collaborated with faculty to design assignments and provided comprehensive feedback to students.
- •Led review sessions, helping students improve their understanding of Python concepts.
Projects
A fully modular, gpu accelerated, deep learning framework made from scratch in C++, CUDA, and OpenCL. (current project linked is an old C build!)
Key Features
- Modular deep learning model builder for flexible architecture.
- Integrated MNIST dataset loader for quick training and testing.
- GPU acceleration for faster computations using CUDA and OpenCL.
- Supports object-oriented programming (OOP) for scalable and maintainable code.
- Test-driven development (TDD) for ensuring robust and bug-free code.
Challenges Overcome
- Implemented efficient memory management for GPU acceleration using CUDA and OpenCL.
- Designed a flexible model builder framework to allow easy customization of layers and training configurations.
- Set up GitHub Actions for automated testing and continuous integration to streamline the development process.
Developed a discord sustainability card that promotes sustainable practices. Deployed on Microsoft Azure and utilized Python, Docker, and MongoDB.
Key Features
- Open-source discord bot focused on sustainability awareness and practices
- Point based reward system used to redeem real prizes.
- Deployed on Microsoft Azure for scalability and high availability
- Integrated MongoDB and SQLite3 for managing user data and progress
- Dockerized application for easy deployment and maintenance
Challenges Overcome
- Optimized deployment on Microsoft Azure to handle increasing user load
- Ensured application security and performance during scaling
- Integrated data persistence and efficient querying with MongoDB and SQL
- Collaborated with non-developers to transform their vision into a working product, ensuring that their sustainability goals were accurately captured in the final product.
A realtime feature detection model for the valorant radar system built purely off of synthetic data generation.
Key Features
- 93% precision, 91% recall
- Real time tracking
- Radar feature detection with template matching
- YOLOv8 Computer Vision
- Synthetic Data Generation
Challenges Overcome
- Lack of manually labeled data. (which is why I resorted to synthetic data)
- Preprocessing raw footage into usable data.
Habitizer
A habit-tracking app that helps users build and maintain positive routines through goal setting, progress tracking, and personalized insights. It encourages consistency and motivation by visualizing habits and providing reminders.
Key Features
- Integration testing for ensuring reliability
- Used Gradle for package management and dependencies
- Followed the Single Responsibility Principle (SRP) for better maintainability
- SQL database for persistent user data
Challenges Overcome
- Project management of a team of 6 developers
- Ensured smooth integration testing and functionality
E-book/PDF reader with login authentication for security for homeservers.
Key Features
- Flask-based backend for secure data handling
- RESTful API for managing user data and books
- Designed and developed the frontend using Tailwind, TypeScript, and Vite for a modern, responsive UI
Challenges Overcome
- Implementing login authentication
3D Printed Ergoboard
A fully 3D printed, handwired keyboard, controlled by an arduino pro micro, engineered for ergonomics.
Key Features
- Ergonomic design
- Uses Custom QMK firmware
- TRRS Connection for both halfs
Challenges Overcome
- Hand soldering all 80 switch connections.
- Configuring custom firmware
- Using a CAD software to edit 3D printed parts
My Resume
Feel free to download my resume for a detailed look at my experience and skills!✨
Let's Connect!
kamaniti@ucsd.edu
GitHub
Check out my open source projects
Let's connect professionally
Don't be shy! I'm always excited to connect with fellow developers and tech enthusiasts!٩(◕‿◕｡)۶