A selection of work spanning machine learning, computer vision, hardware design, and robotics.
For a complete list, please visit my GitHub.
Control Algorithms for Vision-Based Navigation
Undergraduate thesis developing predictive drone navigation using Model Predictive Control (MPC).
Integrated YOLOv8 for perception and achieved 25% reduction in trajectory deviation over PID controllers.
Tested in Microsoft AirSim for high-fidelity simulation of autonomous racing scenarios.
LaLigaLens
Machine learning application predicting La Liga football outcomes and simulating league standings.
Trains ensemble models (RandomForest, XGBoost, LightGBM) on historical match data and team statistics
to forecast future seasons with estimated expected goals (xG).
YuLatin: Latin Quote Generator
AI-powered model analyzing emotions from user input to generate contextually relevant Latin quotes.
Built with Hugging Face's roberta-base-go_emotions for sentiment detection, deployed as a full-stack
application using Node.js and Express.
MazeGest: Gesture-Controlled Maze
Interactive maze game leveraging real-time hand gesture recognition via MediaPipe Hands.
Players navigate dynamically generated mazes using hand movements captured through webcam,
with collision-aware gameplay implemented in vanilla JavaScript.
Speculative Out-of-Order RISC-V Processor
Designing and verifying out-of-order processor implementing RV32IM ISA with floating-point
and multiplication extensions. Features include Perceptron branch predictor with BTB,
stride prefetchers, and split load-store queue for optimized performance.
Ray Tracing Hardware Accelerator
Hardware accelerator for ray-triangle intersection achieving 70.8% performance improvement
and 69.2% energy reduction. Implemented as FSM with custom floating-point units integrated
with RISC-V RI5CY core.
Multi-Cycle RISC-V Processor
Designed multi-cycle processor using the RISC-V instruction set and standard CPU data path.
Implemented register-immediate instructions, load/store memory operations, and conditional
branch operations. Verified design using RVFI monitor and SPIKE.
System Verilog Projects on DE10-Lite FPGA
Replicated classical Tetris with NIOS II and SOC using 32-bit modified Harvard RISC-V architecture.
Developed three-stage (fetch, decode, execute) SLC3 processor for LC3 ISA. Hardware implementations
tested on breadboard and synthesized using Quartus Prime.
Single Core Unix x86 Kernel
Implemented device drivers for keyboard, RTC, Terminal, and TUX controller.
Introduced compatibility for Linux shell commands with support for concurrent processes.
Designed paging-based virtual memory, context switching, and round-robin scheduler.
Convolution Neural Network for Image Classification
Optimized forward-pass of convolution layers using NVIDIA CUDA Platform.
Implemented modified LeNet5 architecture for Fashion MNIST dataset.
Used unroll and shared-memory matrix multiply with kernel fusion to maximize performance.
UR3 Robotic Knight's Tour
Implemented Open Knight's Tour on physical board using modified Hamiltonian Path Algorithm.
Integrated UR3 robot control with inverse kinematics and blob detection for precise movement.
Utilized ROS and ROS Bags for real-time execution and synchronized motion control.
Nonlinear Underactuated System
Developed six controllers to stabilize Reaction Wheel Pendulum using optical encoders.
Simulated three-state feedback controller in MATLAB Simulink to stabilize rotor velocity.
Process included system identification, model validation, and design validation.
Superheterodyne AM Receiver
Built receiver on breadboard with RF/IF amplifier, mixer, demodulator, local oscillator,
and audio amplifier. Designed digital filter to demodulate AM signal and replace IF signal
with soundcard-based sampler for software radio implementation.
Sports Market Inefficiency Detection
Designed advanced analytics framework to uncover market inefficiencies and arbitrage opportunities.
Engineered automated Python pipelines to scrape and structure large-scale sports market data
from major platforms for data-driven decision making.
Open Flights Airplane Path Optimization
Developed program to find shortest path between airports using OpenFlights dataset.
Created weighted directed graph using adjacency lists. Implemented both Floyd-Warshall
and A* algorithms for efficient pathfinding.