Seeking Summer 2026 Opportunities

Hi, I'mSelvahini Kamalarajan

Computer Engineering @ University of Waterloo

Delivering human-centered solutions and user-friendly experiences.

View My WorkGet In Touch
Scroll to Explore
explorer_profile.exe

$ Initializing transmission...

ABOUT ME

>> Origin: Earth | Sector: Developer Quadrant | Status: Active

Background

As a second-year ECE student at UWaterloo, I build scalable, innovative solutions at the intersection of software development, AI, machine learning and embedded systems.

Beyond tech, I enjoy learning about philosophy, cognitive science and astrophysics. I published my first short story with Polar Expressions Publishing at age seven. In my free time, I practice calligraphy and read dystopian fiction.

What I'm Working On

Multi-Agent Systems

Exploring Multi-Agent Orchestration and Physical AI with World Foundation Models (WFMs)

Embedded Systems

Gaining hands-on experience with STM32 Nucleo Board and C++ programming

Full-Stack Development

Building scalable web applications with modern frameworks and cloud technologies

AI Engineering

Developing and deploying machine learning models for real-world applications

TECH STACK

Languages

C++CPythonSQLJavaScriptJavaC#HTML/CSS

Frontend

React.jsNext.jsVue.jsThree.jsReact HooksTailwind CSSBootstrapAxios

Backend & Databases

Node.jsExpress.jsFastAPIFlaskPostgreSQLSQLite

AI & Machine Learning

LangGraphLangfuseLiteLLMMicrosoft Agent FrameworkSemantic KernelPyTorchTensorFlowKerasscikit-learnOpenCVNumPyPandasMatplotlibLoraPEFT

Embedded Systems & Firmware

ESP-IDFESP32FreeRTOSSTM32CubeIDEEmbedded C/C++UARTSPII2CPWMWatchdog TimersCMakeFirmware PackagingSemantic Versioning

Hardware & PCB Design

KiCadSchematic DesignPCB DesignPCB FootprintsQuartus PrimeSoldering

Cloud, Infrastructure & DevOps

Azure FunctionsAzure Blob StorageAzure AI FoundryAzure Container AppsAzure Key VaultBicepGitHub ActionsManaged IdentityDockerCI/CDSonatype NexusRelease AutomationArtifact ManagementGit Submodules
My Work

Featured Projects

Summus – Terms & Conditions AI Assistant

Summus – Terms & Conditions AI Assistant

Chrome extension that summarizes website terms and conditions, highlights privacy concerns, and answers user questions via an AI chatbot. Uses a FastAPI backend with transformer models and Ollama Mistral to extract, analyze, and serve content in real time.

JavaScriptHTML/CSSFastAPI+4
AgriVision – Smart Farming Assistant

AgriVision – Smart Farming Assistant

Features a YOLOv8 object detection model for plant disease classification and a GPT chatbot for farming advice. Hardware integration includes an Arduino-based soil moisture sensor and relay control system for automatic irrigation.

ReactGoogle ColabTailwindCSS+9
Hack the Valley X
Schema Sync: GitHub Copilot for Bank Data

Schema Sync: GitHub Copilot for Bank Data

Two banks, two schemas, one click. Uses SBERT embeddings and cosine similarity to map fields, merge mismatched bank datasets, and generate completeness reports, audit logs, and one-click Excel/PPT exports.

React (Next.js)TypeScriptSQLite+6
SnowSense IoT Network – Sidewalk Hazard Detection

SnowSense IoT Network – Sidewalk Hazard Detection

IoT sidewalk hazard detection system. Features Arduino hardware (ultrasonic/temperature sensors, LED alerts, servo barrier) paired with a live Flask dashboard powered by a sigmoid AI Risk Score and LOF anomaly intelligence.

ArduinoHC-SR04DHT11+5
ESP32 Retro Arcade Handheld

ESP32 Retro Arcade Handheld

Self-contained retro arcade handheld built on an ESP32 microcontroller in C using FreeRTOS. Features a TFT display, custom joystick/button input, PWM buzzer audio state machine, and 5 built-in games (Pong, Breakout, Flappy Bird, Space Invaders, Maze).

ESP32CFreeRTOS+3
Pneumonia Classifier Web App

Pneumonia Classifier Web App

A deep learning model deployed with Streamlit to classify chest X-ray images as normal or pneumonia-affected. Trained on 5,000+ X-ray images to assist with early pneumonia detection and reliable medical diagnosis.

KerasNumPyStreamlit+2
Intelligent Traffic Flow Optimization

Intelligent Traffic Flow Optimization

Developed an AI model to predict traffic flow using the METR-LA dataset. Performed SHAP analysis to interpret feature importance. Created data visualizations to explore traffic trends and built both linear regression and neural network models for prediction.

StreamlitPythonPandas+6
Bad Ice Cream Game

Bad Ice Cream Game

Recreated the two-player Bad Ice Cream arcade-style game using Java and core OOP principles including inheritance, encapsulation, and polymorphism. Visualized with UML diagrams.

JavaEclipseOOP+1
Medical Pager Application

Medical Pager Application

Built a full-stack real-time chat application for medical communication using Stream APIs. Integrated user authentication and session management with Express and Axios.

ReactTailwindCSSJavaScript+4
Plastic Frenzy – Ocean Cleanup Game

Plastic Frenzy – Ocean Cleanup Game

An underwater adventure game built in Unity to promote plastic pollution awareness. Players collect virtual ocean waste, with real-world cleanup tied to engagement.

UnityC#
AssistWave – Medical Assistive Device

AssistWave – Medical Assistive Device

Ultrasonic sensor tracks hand movements within distance thresholds. An STM32 microcontroller transmits data via UART in Single Wire Half-Duplex Mode to a receiving Nucleo board. Output actions include LEDs, fans, and servo-controlled doors.

STM32CubeIDEUML and I/O DiagramsC+++1
Simulated FPGA Traffic Light Controller

Simulated FPGA Traffic Light Controller

Designed and simulated a traffic light controller on an Altera MAX-10 FPGA using a Moore State machine. Implemented combinational and sequential logic using behavioural VHDL.

VHDLIntel Quartus Prime
Fine-Tuned Sentiment Analysis (LoRA/PEFT)

Fine-Tuned Sentiment Analysis (LoRA/PEFT)

Fine-tuned DistilBERT on IMDb using Hugging Face Transformers and the Trainer API with CUDA-accelerated PyTorch. Achieved 89.19% accuracy on the test split.

DistilBERTLoRAPEFT+3
Simulated FPGA Home Energy Monitor

Simulated FPGA Home Energy Monitor

Designed and simulated an HVAC system on an Altera MAX-10 FPGA using VHDL and D flip-flops. Utilized Intel Quartus Prime to implement magnitude comparators, multiplexers, adders and shift registers.

VHDLIntel Quartus Prime
Smart Navigation Cane for Visually Impaired

Smart Navigation Cane for Visually Impaired

ESP32-CAM captures live images over WiFi to a FastAPI backend. YOLOv11 (COCO dataset) identifies obstacles and returns audio feedback via Google TTS. Results displayed in a React web app.

ArduinoESP32-CAMGoogle TTS+3
Hack the North 2025
Helio – Early AI Wildfire Detection

Helio – Early AI Wildfire Detection

YOLOv8 trained on satellite imagery detects wildfire events. An autonomous SparkFun RedBoard rover navigates affected regions. Features a Canadian wildfire map and Twilio-powered resident alerts.

SparkFun RedBoardMotor DriverPyTorch+4
Personal Portfolio v1 – Space-Themed

Personal Portfolio v1 – Space-Themed

Space-themed personal portfolio featuring Framer Motion animations, interactive Spline 3D models and a responsive UI with Radix UI.

React.jsTailwindCSSFramer Motion+4
Personal Portfolio v2 – Cinematic

Personal Portfolio v2 – Cinematic

A cinematic developer portfolio inspired by Interstellar, featuring 3D elements, parallax backgrounds and GPU-optimized animations. Includes a blog page with dynamic slug-based routing.

TypeScriptFramer MotionThree.js+3
ESP32 Automated Irrigation System

ESP32 Automated Irrigation System

Automated closed-loop irrigation system utilizing ESP32 microcontrollers, moisture sensors, and dynamic watering thresholds. (Currently In Development)

ESP32C++Sensors+1
Bracket Bot Sock Matchmaker
In Progress

Bracket Bot Sock Matchmaker

Automated closed-loop irrigation system utilizing ESP32 microcontrollers, moisture sensors, and dynamic watering thresholds. (Currently In Development)

ESP32C++Sensors+1

WORK EXPERIENCE

Community Impact

LEADERSHIP & MENTORSHIP

HiveMind Program

Mentoring youth in science, math, technology and engineering subjects

15+ students tutored

WiE Catalyst Conference

Facilitated hands-on technical workshops and panels

50+ attendees inspired

STEM Nights

Bringing science and technology excitement to schools across KW

1800+ students reached

GenAI Booths

Sparked insightful conversations about AI technology on campus

700+ participants engaged
17th Annual

UW Teaching & Learning Conference

Co-authored a panel proposal and joined as a student speaker, discussing the evolving role of AI in higher education with faculty and peers — a continuation of the conversations started at the GenAI Booths.

Role
Student Panelist
Format
Panel Discussion
Reach
40+ Engaged

PERSONAL LEARNINGS

RECENT WRITINGS

LATEST ENTRY

Fine-Tuned Sentiment Analysis with LoRA and DistilBERT

Movie review sentiment analysis using a fine-tuned DistilBERT transformer model. The model is trained and evaluated on the IMDb dataset with Hugging Face’s Transformers library and accelerated with CUDA on an NVIDIA GeForce RTX 4060 Laptop GPU. Fine-tuning is performed using Low-Rank Adaptation (LoRA) with Parameter-Efficient Fine-Tuning (PEFT), integrated into Hugging Face’s Trainer API. Inference is deployed through a Streamlit web application displaying predicted sentiment labels (positive/negative), confidence scores and runtime device information (CPU/GPU).

Jan 4, 20254 min read
Read Full Article

The Impact of AI-driven Government Surveillance on Democracy and Liberty

Imagine a world where AI is weaponized as a mass surveillance tool and knows us better than we know ourselves. This paper will explore the harmful applications of facial recognition technologies and their potential for accelerating totalitarianism through an Orwellian surveillance system.

Nov 2710 min read
Read Post

Harnessing Convolutional Neural Networks to Discover Exoplanets

Somewhere in the universe, there are distant Earth-like planets that could harbour life. These exoplanets can help us understand the origin, evolution and future of life and how solar systems and planets are formed. But a question arises: how can we find these distant worlds? The answer lies in convolutional neural networks.

Oct 104 min read
Read Post
Explore All Articles