All Projects

A comprehensive collection of my professional, personal, and academic work spanning AI architectures, full-stack systems, and engineering pipelines.

Multispectral Object Detection in Traffic Scenarios

Conventional object detection systems in autonomous vehicles rely primarily on RGB camera data. These systems often struggle under adverse conditions including: • Low-light environments • Fog or rain • Poor contrast between objects and background Such limitations can reduce detection accuracy and impact safety in real-world traffic environments.

PythonPyTorchComputer Vision+3

AI Voice Agent

Many organizations rely on manual phone handling for routine inquiries, resulting in delays, inefficient communication, and increased workload for staff. Traditional call systems lack automation capabilities and cannot intelligently interpret caller intent.

Twilio APINode.jsFastRTC+2

CAEDM Website Redevelopment & Azure Cloud Migration

The legacy website infrastructure limited scalability, performance, and flexibility for future digital services. Updating content and integrating new systems was increasingly difficult due to outdated architecture.

WordPressMicrosoft AzureSQL Server+2

Deploying OpenVLA on SiMa.ai MLSoC

Modern multimodal AI models such as OpenVLA are typically developed using frameworks like PyTorch and designed to run on powerful GPU infrastructures. However, deploying these models on edge AI hardware presents several challenges: • Unsupported PyTorch operations on specialized accelerators • Dynamic tensor shapes incompatible with static computation graphs • High computational cost of large transformer architectures • Hardware constraints on memory usage and power consumption Without significant architectural adaptation, these models cannot be deployed efficiently on edge devices used in robotics or embedded AI systems.

PyTorchONNX RuntimeModel Quantization+6

CoachMO – AI Question Generation System

Traditional endurance coaching platforms rely heavily on manually created question sets or static workout descriptions. This approach introduces several limitations: • Manual question generation requires significant time and effort • Static questions fail to adapt to diverse workout structures • Lack of contextual questioning reduces athlete engagement Existing systems such as Mottiv achieved only about 75% question relevance accuracy, leaving room for improvement in automated coaching systems.

PythonLangChainOpenAI GPT-3.5+3

Tissue-Cut Visualization (BioVis 2024 Challenge)

Spatial transcriptomics datasets contain large volumes of multidimensional biological information that combine gene expression values with spatial tissue coordinates. Traditional data visualization tools struggle to effectively render and explore such datasets interactively. Researchers often face challenges including: • Difficulty linking gene expression data with spatial tissue locations • Lack of interactive tools for exploring multimodal biological datasets • Performance limitations when visualizing high-resolution spatial data These limitations hinder effective biological interpretation of spatial omics datasets.

ViteTailwindCSSThree.js+3

Custom OCR System with PaddleOCR

Traditional OCR systems often struggle with real-world industrial images due to noise, blur, low contrast, and irregular text layouts. Generic OCR models are not optimized for domain-specific datasets and often produce inaccurate results.

PythonPaddleOCROpenCV+2

EuTACT Digital Transformation Platform

Many small organizations lack the technical infrastructure required to adopt modern digital solutions. Existing websites are often static, difficult to maintain, and lack integrations with automation tools or AI services. Deploying scalable digital systems also requires significant development time and expertise.

Next.jsTypeScriptTailwindCSS+3

AI-Powered Mass Finder Dashboard

Mass schedules across the Archdiocese were distributed across multiple parishes and formatted inconsistently, making discovery inefficient for users. Traditional website navigation required manual browsing through individual parish pages, creating friction for users trying to locate nearby Mass services quickly. The lack of centralized data processing, geolocation awareness, and intelligent filtering made it difficult to surface the most relevant Mass options in real time.

WordPress REST APIsPHPSQL Server (EASI Views)+4

Google Sheet as Dynamic Database

Updating structured website data traditionally required code changes or direct database access, which created operational bottlenecks and increased reliance on developers for simple updates.

PHPWordPress REST APIGoogle Sheets (Published CSV)+1