Storm Schutte

Developed a novel weapon-detection model leveraging deep learning to contribute towards public safety, identifying nine different types of weapons with an F1 score of 83.3%. The model was trained on a dataset of 52,000 weapon images, expanded through data-augmentation techniques.

The model was architected around Convolutional Neural Networks (CNNs), a proven approach for image-recognition tasks. It was tested across diverse mediums (CCTV footage, movie scenes, still pictures, and live video) to validate real-world performance.

The objective was to strengthen security in public spaces by enabling real-time weapon detection, demonstrating the role AI can play in bolstering security through intelligent surveillance.

Developed an audio classification model to address the rise in school shootings in the USA, focused on recognizing and differentiating sounds, with primary emphasis on detecting gunshots and alerting designated individuals by email.

The model was trained on a large dataset of audio featuring ambient noise, construction sounds, barking, sirens, horns, and gunshots. Built on a CNN architecture with a ReLU activation function and the Adam optimizer, using TensorFlow, Keras, and Librosa.

The system includes a real-time alert mechanism that automatically sends emails when emergency sounds are detected, demonstrating a practical path to faster emergency response.

Built a computer-vision pipeline to detect and count potholes on city roads in real time, for a municipal client that needed an automated way to quantify road-surface damage across regions.

A forward-facing camera mounted on vehicles continuously captured road footage. The video stream was processed by a custom deep-learning model, a CNN optimized for object detection in challenging outdoor conditions with variable lighting and road surfaces, architected and trained from scratch.

The workflow covered data acquisition and manual bounding-box annotation, model development with transfer learning, and a video-processing pipeline that ingests live footage, runs frame-by-frame detection, and outputs pothole counts with minimal latency.

The system achieved 98% detection accuracy with near-zero false positives on unseen test data and field trials, significantly cutting manual inspection requirements and enabling data-driven infrastructure maintenance.

Built a smart-home IoT device that integrates environmental sensing, user interaction, and health monitoring into one system, developed on a Raspberry Pi running Ubuntu Server as the central hub for data collection, processing, and device orchestration.

Voice interaction: distributed microphone arrays capture audio in real time; an NLP pipeline interprets commands, connected to OpenAI's language models for conversational responses and task execution.

Home automation: Wi-Fi modules integrate lighting, automatic curtains, speakers, and alarm systems, triggering environmental adjustments or notifications.

Health monitoring (prototype): a biometric module tracks breathing, heart rate, perspiration, and movement, aimed at supporting people with chronic conditions through real-time wellness alerts.

Architecture: a modular IoT framework combining real-time data ingestion, event-driven processing, and edge computing on the Raspberry Pi, with cloud-assisted AI inference for language understanding.

The system is deployed in several homes today, providing automated control and conversational assistance; the health-monitoring component remains at prototype stage.

Since 2007 I've developed and deployed 100+ websites spanning corporate, non-profit, educational, and personal domains. It covers the full lifecycle from frontend design to backend architecture, infrastructure, and deployment at scale.

Frontend: HTML, CSS, React, and modern JS frameworks. Backend: PHP, Python (Django/Flask), Node.js, and custom APIs. Cloud & hosting: AWS (Elastic Beanstalk, EC2, Amplify), Google Cloud Platform, Cloudflare, PythonAnywhere, Heroku, plus custom deployments on Microsoft servers and self-hosted environments.

What I've built: scalable corporate sites for 100+-employee organizations; CRM and enterprise platforms including academy systems and stock-trading platforms with real-time tracking and scoring; international, multi-language community platforms; and pro-bono sites for non-profits.

Applied cloud-native practices (CI/CD, load balancing, distributed hosting) and DevOps principles for deployment automation, giving me deep experience in scalable architecture, cross-platform deployment, and secure, data-driven applications.

Designed and deployed a fully immersive VR application delivering 360° video experiences for stress relief and meditation, built with Unity and packaged for iOS, Android, and Windows.

Core features: high-resolution 360° environments with interactive layers; audio-visual synchronization combining calming music, guided meditation, and custom VFX; and in-session interactivity so users can personalize their experience.

Technical implementation: captured and optimized 360° footage for seamless Unity playback, engineered a single codebase for builds across the Apple App Store, Google Play, and Windows, and managed the full lifecycle from design through app-store publishing.

Delivered a production-ready app adopted by 10,000+ users, executed as a two-person team spanning Unity development, VR video pipelines, and app-store publishing.

Explored machine learning and deep learning to detect, classify, and recommend interactions with 3D objects inside a metaverse environment, building a system that understands 3D structure well enough to power recommendation engines and virtual-environment optimization.

Worked extensively with 3D mesh data (vertices, edges, faces), addressing terabytes of geometric data through dimensionality reduction and efficient preprocessing. Compared classical ML (SVMs, random forests) against deep architectures (CNNs for voxel data, PointNet-style models for point clouds), then extended classification into content-based recommendations.

Designed custom training workflows to optimize GPU utilization and reduce cloud compute costs, using batch sampling and sparse data representations to make deep learning feasible at terabyte scale.

Delivered a working 3D detection and classification pipeline, and showed that well-engineered classical ML can rival deep learning when paired with strong feature engineering, a meaningful cost saving for large-scale 3D data projects.

Tag9 is an application for creating, sharing, and collecting data through customizable forms, combining ease of use, versatility, and robust security for businesses and organizations.

Key features: flexible form creation with a user-friendly interface, efficient data sharing via QR codes or tokens, advanced security for data collection, and integration with existing software systems.

Built on Django, backed by PostgreSQL, and enhanced with JavaScript for an intuitive front end, with data-security measures designed to meet privacy and compliance standards.

PEIMS is an academy management system covering everything from class management to customer financial tracking, in-class management, and staff/student administration, designed to fit as many academy types as possible.

The platform focuses on the essential features academies actually need: class management, customer financial management, in-class management, and staff/student administration.

Built with Django for a scalable solution suited to educational institutions seeking efficient management tools.

Built an intelligent testing system for the Korean Government that reworks traditional examination methods, using algorithms that dynamically adjust questions based on individual performance throughout the exam.

The system analyzes test-taker responses in real time, adapting question difficulty and type to match the candidate's skill level and learning progress.

This adaptive approach produces more accurate assessments and a more personalized exam experience, a meaningful step forward in educational assessment technology.

A cross-platform mobile app built for both Android and iOS, designed for a business that needed to make audio files accessible to its members through a dedicated app.

Features a custom audio player, secure content delivery, and user authentication so only authorized members can access the audio content.

Delivered a professional cross-platform solution tailored to the client's requirement for secure member-only audio distribution.

As team lead, I designed and deployed a retrieval-augmented semantic search system for an automotive CRM platform, working with a professor, three developers, and government stakeholders.

Problem: the dataset held millions of records across brands, models, parts, and regional variants. Queries were highly ambiguous ("Sonata 2.0 vs Sonata Hybrid," "SUV with 4-wheel drive") and often cross-lingual (Korean/English). Keyword search returned irrelevant results, with baseline accuracy around 26%.

Approach: built an NLP preprocessing pipeline with spaCy and custom entity extractors; benchmarked open-source embeddings (Sentence-BERT, multilingual MiniLM) against OpenAI's large embedding model, which proved superior for multilinguality without needing per-language models.

Vector search: implemented Weaviate for high-dimensional nearest-neighbor search, with re-ranking heuristics to reduce near-duplicate results. Evaluated Neo4j for hierarchical relationships but excluded it: complexity outweighed the marginal benefit.

Results: increased search accuracy from 26% to 98%, validated against a labeled benchmark of real CRM queries, without adding a separate translation pipeline, supporting a government-backed initiative in Korea.

Key learning: embeddings generalize across languages better than fine-tuned monolingual models when the goal is scalable retrieval, and vector databases like Weaviate are production-ready for semantic CRM at industry scale.

Led development of a large-scale video management and review application for schools, organizations, and creative communities: users could upload videos and leave reviews, while directors and creators engaged directly with their audience.

Recommendation engine: collaborative-filtering foundation for discovering videos from viewing history, engagement patterns, and community trends.

Sentiment analysis: prototyped NLP models to score comment sentiment and better surface high-quality content.

Engagement analytics: dashboards tracking watch time, likes, and review activity for creators and competition organizers.

Content moderation: experimented with ML-based moderation to flag inappropriate content or abuse at scale.

Architecture: hosted on AWS EC2 to absorb spikes during competitions and live events, with storage tuned for large volumes of user-generated video and real-time interaction.

Successfully launched and scaled the platform, which built a dedicated following and continues to run under new management, pairing a simple, user-friendly interface with ML-driven engagement and discovery.

Since 2018 I've built seven React applications: from complex data-driven platforms with dynamic inputs and real-time updates, to Progressive Web Apps (PWAs) that run seamlessly on mobile, mostly styled with Tailwind CSS.

Framework evolution: since 2022, all new projects use Next.js for its performance and developer experience; PWA builds give users an installable, app-like experience straight from the browser.

Project types: data-driven platforms with dynamic forms and interactive dashboards; installable PWAs across mobile and desktop; and real-time applications with live updates and notifications.

Technical depth: React, Next.js, JS ES6+, and TypeScript for type safety; Tailwind CSS and CSS-in-JS for styling; Redux, Context API, and hooks for state; code-splitting and lazy-loading for performance.

Combined hands-on delivery with structured study of React's component model and state-management patterns, giving me both a strong theoretical foundation and the ability to ship production-ready apps.

Designed, deployed, and managed the entire IT infrastructure for a team of 47 employees at Switch Air, a Microsoft Server-based environment, with direct responsibility for system administration, network security, and end-user support.

Server & directory services: Active Directory for centralized authentication, group-policy management, and role-based access control.

Email & collaboration: company-wide email via Microsoft Exchange (on-premise and cloud), with secure communication and spam filtering.

Networking & security: domain management, firewall configuration, VPN access, and endpoint security.

Automation: automated PC updates, patch management, and backups, cutting vulnerabilities and downtime.

Built a reliable, secure IT ecosystem that improved business continuity, reduced downtime, and established practices that scaled with company growth.

Designed, built, and managed enterprise data centers and server rooms from the ground up, engineering the environmental and safety systems to the industry standards of the time: precision temperature and humidity control, fire detection and suppression, and large-scale backup power spanning industrial generators and battery stacks, later extended with solar power for greener, more resilient operations.

Infrastructure management evolved alongside the industry, from administering bare-metal servers and Microsoft Server environments to modern containerized and orchestrated infrastructure running on Kubernetes.

Today, manage a server room supporting multiple companies, instrumented with modern IoT sensors and controls that monitor and manage every aspect of the facility in real time, including temperature, power, and fire safety, continuing a practice of hands-on data center reliability engineering that spans nearly two decades.

Presents an IoT security system leveraging machine learning for real-time threat detection in smart environments, using Raspberry Pi devices as edge computing nodes, Google Vertex AI for threat analysis, and AWS for scalable data processing and storage.

Demonstrates significant improvements in detecting potential security threats, particularly identifying individuals carrying weapons in real-time scenarios, contributing to the field of edge AI and smart security systems.

Examines how the COVID-19 pandemic accelerated the digitalisation of education and reshaped K-Pop-driven migration and study patterns to Korea, looking at how online learning and remote engagement changed the pathways international students and K-Pop enthusiasts take to study and live in Korea.

Presents deep-learning approaches for medical image segmentation in cancer diagnosis and treatment planning, focused on convolutional neural network architectures for accurate breast-cancer detection and segmentation.

The proposed methods show measurable improvements in diagnostic accuracy and processing speed, supporting earlier detection and more effective treatment planning in clinical settings.

Contributed a chapter on optimizing educational curricula with large language models to this edited volume, which explores Generative AI's role in education, spanning curriculum development, personalized learning, assessment, and special education.

Shows how LLMs can create adaptive, personalized learning experiences, with a framework for integrating them into educational systems to improve student engagement and learning outcomes.

Explores the intersection of big-data analytics and deep learning for immersive 3D environments in the metaverse, presenting approaches for handling large-scale data processing in virtual-world creation and management.

Shows how ML techniques can optimize 3D object rendering, user interaction, and data flow in virtual environments, contributing to more realistic, responsive metaverse experiences.