Projects

Delve into a collection of my revent notable projects, each representing a fusion of creativity and skill.

Overview

NEW

Exechon Algorithm

2025

Learning-assisted Drifting

2024

Video Panoptic Segmentation

2024

Perception-enabled Pure Pursuit

2023

High-performance Control Design

2023

Policy Learning for an Unbalanced Disc

2023

U-Net Semantic Segmentation

2023

Autonomous Multi-Drone Coordination

2023

3D Point Cloud Filtering

2023

Anomaly Detection with Autoencoders

2023

Exercise Activity Detector

2022

Autonomous Drone Navigation

2021

NEW

2025

Exechon Algorithm

Python

SQL

ETL

API Integration

Optimization

Sentiment Analysis

small-team personal project

Exechon is a modular, layered analytics platform that turns raw SEC 13F filings, stock financial data and real-time news into insight-ready data. A rich CLI and interactive visualizations let professionals explore “what-if” scenarios, stress-test strategies, fine-tune simulations and parameters, and compare performance to similar benchmarks. Its clean architecture, smart storage layer, and extensible design make Exechon a single, scalable hub for data-driven investment research.

Efficient Frontier Allocation

Parameter Optimization

Buy-and-hold Simulation

Buy-side Simulation with rebalancing

2024

Learning-assisted Drifting for F1Tenth Autonomous Race Cars

Python

Matlab

ROS

Optimization

Machine Learning

Control Systems

Associated with TU/e (MSc Thesis)

Developed an advanced control framework for autonomous drifting using small-scale F1Tenth vehicles, integrating automated gain tuning and Gaussian Process-based model augmentation to enhance control accuracy and path tracking in complex trajectories.

2023

Depth-Aware Video Panoptic Segmentation & Monocular Depth Estimation

Python

Deep Learning

Neural Networks

High Performance Computing

Associated with TU/e

Developed a multimodal deep neural network for simultaneous panoptic segmentation and monocular depth estimation from still images, enhancing depth-aware scene understanding.

Original Color Image

Generated Semantic Mask

Generated Instance Mask

Generated Depth Map

Enhanced Gaussian Centers

2023

Semantic Segmentation using a UNet-based Architecture

Python

Deep Learning

Generative AI

Encoders

Associated with TU/e

Developed a robust semantic segmentation model for urban street scenes using a U-Net-based architecture, addressing challenges in class imbalance, image quality degradation, and generalization. Enhanced performance through data augmentation, external datasets, transfer learning, and adaptive class weighting, demonstrating improved segmentation on the Cityscapes dataset.

2023

Perception-enabled Pure Pursuit for Small Scale Racing Cars

Python

Deep Learning

Object Detection

ROS

Controller

Reinforcement Learning

Associated with TU/e

Developed an autonomous agent for small-scale racing cars to navigate unknown racetracks using object detection for cones and a 2D environmental map. Implemented and modified a geometric controller for improved adaptability and speed while conducting a theoretical analysis of reinforcement learning for decision-making.

F1Tenth Car

Racing Map Example

2023

High-Performance Control Design for a Fourth-Order Rotational Motion Setup

Matlab

Simulink

Control Theory

Frequency Analysis

Associated with TU/e

Designed and implemented a high-performance control system for a fourth-order rotational motion setup, significantly improving trajectory tracking accuracy using advanced feedback and feedforward techniques.

2023

Policy Learning for an Unbalanced Disc

Python

Model Identification

Probabilistic Machine Learning

Reinforcement Learning

Associated with TU/e

Developed a control policy for an unbalanced disc pendulum system using Gaussian Processes, Neural Networks, and Reinforcement Learning, enabling successful swing-up and stabilization while extending control to multiple target positions.

2023

Autonomous Multi-Drone Coordination for Wildfire Detection and Suppression

MATLAB

Control Theory

Flocking

Multi-agent Systems

LTL Specifications

Associated with TU/e

Developed and implemented autonomous control strategies for a swarm of firefighting drones, focusing on fire detection, monitoring, and suppression in dynamic environments. Designed and linearized UAV motion models, synthesized controllers for safe navigation, and implemented formation and leader-follower flocking control for coordinated drone movement. Optimized fire suppression efficiency by simulating bushfire behavior and adapting drone formations, demonstrating improved coverage and responsiveness. The project successfully integrated multi-agent coordination and control theory, enhancing UAV autonomy in disaster response scenarios.

2023

3D Point Cloud Filtering

Python

ROS

Point Clouds

Filters

Associated with TU/e

Developed a custom Robot Operating System (ROS) node to filter a noisy cumulative point cloud, enhancing data quality for improved perception and navigation in robotic applications.

Unfiltered (Top Down View)

Filtered (Top Down View)

Unfiltered (Back Left View)

Filtered (Back Left View)

2022

Anomaly Detection with Autoencoders

Python

Deep Learning

CNNs

Unsupervised Learning

Generative AI

Associated with TU/e

Implemented and analyzed an autoencoder from scratch for anomaly detection on the MNIST dataset, leveraging unsupervised learning to detect deviations from normal digit patterns.

MNIST dataset (Original, Noisy and Reconstructed)

2022

Exercise Activity Detector

Python

Unsupervised Learning

Time-series Analysis

Sensor Fusion

Associated with TU/e

Developed an Unsupervised Machine Learning-Based Activity Detector by applying K-means clustering and probabilistic modeling to sensor recordings. Preprocessed data, explored deterministic approaches for clustering, and implemented a detection system capable of identifying distinct activities based on sensor inputs.

2021

Autonomous Drone Navigation for Landmark Position Estimation using Reinforcement Learning

Python

Reinforcement Learning

ROS

Neural Networks

Unsupervised Learning

Associated with TUC

Developed a mapless UAV autonomous navigation system using deep reinforcement learning to navigate unknown 3D environments, detect ArUco markers, and avoid procedurally generated obstacles. Evaluated performance across varying environment complexities, demonstrating the potential of DRL for autonomous UAV missions.

Before Training

After Training

Obstacle Family

Marker Family

More projects

For additional projects, check out my LinkedIn profile and coding repositories below.