RecyKool — CSUN Senior Design Project Website

Team & Roles

Group members, responsibilities, and profiles

Project Lead

Johnathan Aguilar

Agentic AI Assistant Backend Developer

GitHub • LinkedIn

Core Dev

Denver Cude

VLM / Agentic AI Assistant Frontend Developer

GitHub • LinkedIn

Core Dev

Alexander Boutselis

Digital Twin Developer

GitHub • LinkedIn

Core Dev

Andy Ruiz

Digital Twin Developer

GitHub • LinkedIn

Core Dev

Angel Cortes-Gildo

Human Gesture Recognition / VLM Developer

GitHub • LinkedIn

Core Dev

David Sterin

Material Characterization / VLM Developer

GitHub • LinkedIn

What RecyKool Does

Project Overview

RecyKool is a multimodal AI system designed to support operations within Waste Management’s conveyor sorting environment. The platform is organized into three core components. The AI Assistant provides a conversational interface backed by EOD reports and equipment manuals, allowing users to query system behavior and receive grounded guidance. A vision-language model (VLM) handles material characterization and flagging, including detection of jam-causing items and recognition of unsafe worker behavior. A digital twin component models the conveyor system, enabling simulation and providing additional context for understanding real-time conditions and operational issues.

Features

Multi Modal AI Assistant (KG + MRAG)

Supports text and image-based querying, connects user requests to retrieved documentation, and uses knowledge graph reasoning to provide more grounded operational guidance.

Digital Twin

Provides an Omniverse-based facility simulation concept for understanding conveyor activity, camera input, and real-time monitoring in a visual operational context.

Jamming Object Detection (VLM)

Uses a vision-language model approach to identify materials that may cause jams or disruptions in the sorting line.

Human Safety / Behavior Modeling

Explores unsafe action detection using real and synthetic data to support earlier awareness of hazardous worker behavior.

Jump to Demo

Skills

Capabilities developed during the project

AI / Vision-Language Models

Model development, Multimodal Reasoning, OCR, RAG, and Knowledge Graph Workflows

Applied multimodal AI techniques including vision-language models, retrieval-based reasoning, and structured knowledge representations to generate grounded explanations.

Digital Twin Systems

Facility Simulation, Camera Integration, Physics Testing, and System Modeling

Developed and interacted with a real-time Omniverse digital twin of a conveyor belt system to simulate operations and visualize system behavior.

Frontend Development

Spatial interface design, Immersive UI testing, and User-facing AI Tools

Built and tested an immersive frontend experience for Apple Vision Pro, focusing on spatial UI interaction and usability.

Project Stack

Technologies Used

The main tools, models, platforms, and frameworks used across the AI assistant, detection models, digital twin, synthetic data pipeline, and XR interface.