Abheek Pradhan

Computer Engineering Student | Full stack and embedded systems

Abheek Pradhan

About Me

I am a driven and well-rounded senior at Texas State University. I have a knack for optimization and enjoy tackling challenging problems in full-stack/back-end development, machine learning and embedded systems.

Outside of programming, I enjoy cooking, gardening, reading, exercising, and various board sports.

Quick Facts

  • Computer Science Student
  • Embedded Systems
  • Full Stack Development
  • Problem Solver

Work Experience

Software Engineering Intern

Toshiba International Corporation

May 2025 - September 2025

Remote

  • Developed STM32 multi-core FreeRTOS (real time operating system) firmware and TouchGFX UI for medium voltage drives on ARM/x64 architectures; improved C backend performance by 25% through task priority optimization and stack size tuning using native FreeRTOS and CMSIS-RTOS2 APIs (C, C++).
  • Built automated testing infrastructure using Python, Bash, and Ruby for 10,000+ parameter validation on hardware and simulators; created CI/CD pipelines for all tests (unit, integration, HIL (Hardware in Loop)).
  • Engineered RAG based AI agent for defect detection using Azure Copilot to parse CAD drawings from Oracle database, achieving 90% accuracy in identifying manufacturing defects
CC++STM32FreeRTOSTouchGFXPythonBashRubyAzureOracle

Machine Learning Engineer

Texas State University

December 2024 - Present

San Marcos, TX

  • Engineered ML backend / full-stack systems for Alzheimer's research project. Deployed scalable ML backend using Hugging Face Spaces with Docker containerization and AWS services (S3, API Gateway, DynamoDB, CloudWatch).
  • Fine-tuned Vision Transformer and MASK R-CNN models for object detection / agentic pipeline; led comprehensive dataset creation with systematic annotation protocols, achieving 98% precision through data balancing and augmentation
  • Optimized inference via multi-threading, RGB analysis, and parallel processing across Python, Java, and React Native platforms
PythonPyTorchHugging FaceDockerAWSVision TransformersMASK R-CNNReact Native

Research Assistant

Texas State University

August 2024 - Present

San Marcos, TX

  • Collaborated with PhD students led by Dr. Anne Ngu to assist with training multi modal transformer models and overall knowledge distillation pipeline for fall detection on edge devices with cross modal learning funded by NSF.
  • Responsible for deploying Pytorch / Tensorflow Transformer models to edge (Android) via conversion to Tensorflow lite using pre and post training quantization, achieving <1% accuracy loss and 1.5x-2x battery improvement
  • Built multi-modal data pipeline with RESTful APIs and PostgreSQL database; implemented signal processing algorithms (Butterworth, Kalman filters, sensor fusion) to clean data for knowledge distillation achieving 5% F1 score increase
  • Optimized inference on low-resource hardware (e.g. smartwatch and phone) via multi-threading, batch-processing, refactoring incompatible model operations, and hardware-acceleration (GPU, NPU) resulting in 3x speedup in Java, Kotlin android app.
PythonPyTorchTensorFlowAndroidJavaKotlinPostgreSQLSignal Processing

Featured Projects

A selection of my recent work showcasing my skills in full-stack development, embedded systems, and AI/ML

FPGA-Optimized Facial Recognition YouTube thumbnail

FPGA-Optimized Facial Recognition

TLDR: YouTube

Built hardware-accelerated facial recognition system on Kria KV260 SoC using INT8 quantization, custom Vitis AI kernels, zero-copy DMA architecture, and parallel DPU inference for embedded Linux deployment.

CC++Embedded LinuxYoctoPyTorchFPGAVitis AICUDA
Sensor Fusion for Human Activity Recognition thumbnail

Sensor Fusion for Human Activity Recognition

Developed sensor-fusion pipeline for fall detection using accelerometer/gyroscope data with Complementary, Madgwick, Mahony, and EKF orientation filters optimized for real-time edge computing.

PythonSignal ProcessingKalman FiltersEKFNumPyTime Series AnalysisEdge Computing
3D Skeleton Reconstruction from Video thumbnail

3D Skeleton Reconstruction from Video

Implemented GAST-NET deep learning architecture for 3D human skeletal joint reconstruction from 2D video using computer vision and PyTorch for motion analysis applications.

PythonPyTorchComputer VisionDeep Learning3D ReconstructionOpenCV
Time Series Orientation Estimation thumbnail

Time Series Orientation Estimation

Implemented Kalman, Extended Kalman Filter, and complementary filters for IMU-based orientation estimation and sensor alignment optimized for real-time fall detection on embedded devices.

PythonKalman FiltersIMU ProcessingSignal ProcessingNumPySciPyEdge Computing
Smartphone-Based Fall Detection thumbnail

Smartphone-Based Fall Detection

Developed real-time fall detection system using smartphone IMU sensors with machine learning activity classification and TensorFlow Lite for resource-constrained mobile deployment.

PythonMachine LearningAndroidIMU SensorsTensorFlow LiteEdge ML
Autonomous Chess Bot YouTube thumbnail

Autonomous Chess Bot

TLDR: YouTube

Built autonomous chess application integrating Stockfish API with asynchronous multi-threading, WebSocket communication, database caching, and automated board recognition for online gameplay.

C++Stockfish APIMulti-threadingWebSocketsNodeJSDatabase Optimization
Real-Time Audio Transcriber thumbnail

Real-Time Audio Transcriber

Engineered full-stack audio transcription web application with Flask WebSocket backend, React.js/Next.js frontend, hardware-accelerated Whisper AI speech-to-text, and MongoDB session storage.

PythonReactJSNext.jsFlaskWebSocketMongoDBWhisper AI

Education

Bachelor of Science in Computer Science and Engineering

Texas State University

San Marcos, TX

Expected May 2026

Activities and Societies

.EXEACM Member

Skills

Frontend

  • React
  • Next.js
  • TypeScript
  • Tailwind CSS

Backend

  • Node.js
  • Python
  • Express.js
  • PostgreSQL

Tools

  • Git
  • AWS
  • Docker
  • Linux

Awards & Recognition

🏆

Dean's List

Fall 2023

Recognized for outstanding academic performance with a GPA above 3.5.