Joshua
Elkind

Full-featured Unix shell in Rust with process control, job management, pipelines, and redirection. Implemented safe, concurrent process handling leveraging Rust's ownership model. Integrated signal handling, environment variables, command history, and configuration support.

Cross-platform desktop application built with Electron that provides a visual interface for designing, orchestrating, and monitoring AI agent workflows. Features drag-and-drop canvas for building workflows, multi-LLM support with OpenAI and Deimos Router, real-time analytics dashboard with Chart.js, and Databricks Vector Search integration for document processing and retrieval.

Pharmacogenomics risk assessment platform with ensemble ML models achieving 95.8% AUC. Built distributed ETL pipelines with Apache Spark and Airflow processing VCF datasets from AWS S3 to Snowflake. Integrated real-time clinical decision support via .NET Core API and React frontend.

Full-stack algorithmic trading platform enabling users to run Python strategies on millions of rows of historical stock data. Architected secure cloud-native execution with AWS EKS Kubernetes Jobs and Docker. Integrated Apache Spark for large-scale backtesting and statistical analysis.

High-performance ray tracer in Go leveraging goroutines and worker pools for near-linear CPU scaling. Implemented recursive reflections, soft shadows, volumetric effects, and procedural materials. Designed spatial acceleration structures (BVH, Octree) to maximize intersection throughput.

Scalable full-stack tutoring platform on Azure with live video, real-time chat, and resource sharing. Built AI-powered features with LangChain, Chroma vector DB, and RAG for quiz generation and video transcription. Architected caching with Kafka, Elasticsearch, and Redis for performance at scale.

High-performance fractal zoom animation engine in Go with near-linear CPU scaling. Developed dynamic formula parser for custom fractal formulas. Created intelligent zoom system with boundary detection and adaptive quality scoring for infinite zoom sequences.

Convolutional Neural Network built entirely from scratch using NumPy achieving 92.8% test accuracy on Street View House Numbers. Implemented custom 2D convolution, batch normalization, max pooling, and Adam optimizer. Optimized with He initialization and im2col technique for efficient training.

Generative Adversarial Network built from scratch using only NumPy for generating synthetic alphabet letters. Implemented generator and discriminator networks with binary cross-entropy loss and custom backpropagation. Designed custom dataset of 28×28 alphabet glyphs with noise and transformations.

Real-time face detection and recognition system on Raspberry Pi using OpenCV. Implemented custom logic to differentiate known individuals and strangers, triggering email alerts with face image attachments. Optimized frame processing with background subtraction and contour detection.

React-based web platform integrating Google Maps and Places APIs to display nearby aid resources geographically. Implemented dynamic location-based services for food banks, shelters, and mental health clinics. Designed accessible UI with clear messaging for users without location access.

Fast-paced mobile racing game in Unity with multi-lane traffic avoidance gameplay. Implemented dynamic obstacle spawning and smooth player controls. Integrated coin collection mechanics enabling vehicle and parts upgrades for player progression.

Python tool using Selenium to create and play YouTube playlists offline with automatic ad skipping. Used YouTube API to get video details and calculate total playlist duration. Automated browser actions to control YouTube playback and handle ads.