Josh Rabbani
AI/ML EngineerNew York

Josh Rabbani

I build machine learning systems and the software around them.

Computer Science (AI/ML concentration) & Mathematics at Duke University

What I work with

Tap any skill to see the projects where I used it.

AI & ML

Backend

Frontend

Mobile

Selected work

Things I've built

01

Full-stack platform connecting local businesses with student brand ambassadors, streamlining marketing campaigns and microinfluencing by students. Built on Next.js and Supabase, with row-level security in Postgres.

React & Next.jsJavaScriptHTML/CSSPostgreSQLSwiftSQLDatabase DesignReal-time Messaging

Frontend Architecture

  • Next.js Application
  • Atomic Design Components
  • Tailwind CSS Styling
  • Responsive UI Design

Backend Systems

  • Supabase Integration
  • PostgreSQL Database
  • Row-level Security
  • Real-time Analytics

Key Features

  • User Authentication & Profiles
  • Application Tracking System
  • Campaign Management Dashboard
  • Performance Metrics & Analytics

Mobile app

Rippl screenshot 1
02

Full-stack Android application streamlining meal order processing by Duke students and printing receipts for Durham restaurants in a $120K annual revenue business. Built with hardware integration for StarXpand printers, supporting restaurant operations.

KotlinJavaAndroidHardware Integration

Frontend & Mobile

  • Kotlin Android Application
  • Dynamic Fragment Management
  • Minimalistic, Accessible UI
  • Real-time Order Processing

Hardware Integration

  • Star Micronics StarXpand SDK
  • LAN-based Receipt Printing
  • Multi-protocol: Bluetooth & USB
  • Asynchronous Print Operations

Business Impact

  • Powers $120K+ Annual Revenue Business
  • Streamlined Restaurant Operations
  • Real-time Order Management System

Tablet app

Restaurant Order Processing App Screenshot
03

Developed and deployed a web tool within 10 weeks aiming to help 1.2 million veterans navigate insurance options to find the lowest-cost diabetes medications, sponsored by Microsoft and the U.S. Department of Veterans Affairs through Duke Office of Information Technology (OIT) Code+ program.

React & Next.jsRuby on RailsPythonPostgreSQLDockerHTML/CSSData ProcessingAPI Development

Frontend Development

  • Next.js Single-page App
  • Atomic Design Components
  • Wireframed questionnaire for scalability
  • Accessibility-focused UI design

Backend & Data

  • Ruby on Rails API
  • PostgreSQL Database
  • 40GB → 500MB Data Cleaning
  • API layer between the Next.js app and Rails

Project Impact

  • Serves 1.2 Million Veterans
  • Microsoft & VA Sponsored
  • 10-Week Development Timeline

Final presentation

Team

Vet-RX Team Photo
04Open on the App Store!

iMessage Extension that lets users challenge their friends to pushup challenges! The inspiration behind this was that I was feeling too stagnant at home, and the 'If you'll do it, I'll do it!' mentality works on me. Thus, Gimme 10 was born to get me to do more pushups.

SwiftProduction DeploymentPersonal Project

App screenshot

Gimme 10 Demo Screenshot
05

End-to-end plant identification and information system built around Duke University's Sarah P. Duke Gardens. Users upload a plant photo, and the system identifies the species using a fine-tuned ResNet18 image classifier trained on species native to Duke Gardens, then answers natural language questions through a RAG pipeline backed by a Wikipedia-sourced plant knowledge base.

PythonFlaskPyTorchResNet18Computer VisionRAGNLPSentence TransformersPhi-2

Computer Vision

  • ResNet18 fine-tuned classifier
  • 102 plant species from Duke Gardens
  • ~22,000 training images sourced via iNaturalist
  • 65.7% test accuracy on held-out data

RAG & Language System

  • Wikipedia-sourced plant knowledge base
  • sentence-transformers/all-MiniLM-L6-v2 embeddings
  • Cosine similarity retrieval
  • Phi-2 local answer generation on GPU

Project Scope

  • Built as an individual project
  • Combines computer vision, semantic retrieval, and language generation
  • Designed as the foundation for a future app with plant discovery tracking

Repository

Source, notebooks, data layout, and setup for the CS 372 final project: Flask app, classifier, and RAG pipeline.

joshrab/372-Final

Also

Other projects & presentations

Durham Diner Operations Analysis for Efficiency

PDF

Simulated restaurant operations to evaluate efficiency trade-offs using a Poisson arrival model and exponential service time distribution. Analyzed customer flow, wait times, and revenue across 300 simulation runs to determine optimal staffing and table configuration for maximizing profit and satisfaction.

Poisson ProcessesExponential DistributionsProfit ModelingStochastic Simulation

Statistical Analysis Project

Key Takeaways

  • • Modeled customer arrivals as a Poisson process and service times using an exponential distribution
  • • Ran 100 simulations across 3 distinct scenarios: varying chefs, tables, and both
  • • Identified optimal configuration: 2 chefs and 5 tables, balancing near-zero wait time and maximum profit
  • • Quantified downtime, wait time distributions, and service throughput to inform strategic resource allocation

Food Accessibility in the US

Report

Investigated how population density correlates with food insecurity across US states using USDA and census-derived data. Focused on proportions of county populations residing 10+ miles from a supermarket, with emphasis on North Carolina. Used R for data wrangling, mapping, and statistical modeling to reveal geographic and socioeconomic disparities in food access.

Geospatial AnalysisLinear RegressionPopulation DensityR Visualizations

Data Analysis Project

Key Takeaways

  • • Used state-level population and access data from USDA and the US Census via the CORGIS dataset
  • • Created choropleth maps and scatter plots in R to visualize total vs. proportional low-access populations across states
  • • Performed linear regression showing a significant negative correlation between population density and food inaccessibility (p < 0.001)
  • • Highlighted disparities in rural vs. urban states, where sparsely populated states (e.g., North Dakota) show >15% low access

SafeRoute: Durham Safety-Optimized Navigation

Live demo

Full-stack navigation platform built during Hack Duke for Durham, NC. Uses React, Python, and OpenStreetMap to analyze 7.7M historical traffic accidents (2016-2023) and compute safety scores with custom pathfinding for risk-weighted routing.

ReactPythonNetworkXOSMnxGeospatial AnalysisPathfinding

2025 Hack Duke Hackathon

Technical highlights

  • • Python data processing with NetworkX and OSMnx for graph-based routing
  • • Custom pathfinding with risk-weighted route optimization
  • • 7.7M accident records from the Kaggle dataset
  • • Folium map visualization with location presets
  • • Completed in a 24-hour sprint

Room Scout

Pitch

LiDAR-powered room scanning platform for student housing decisions. Won audience favorite in the Duke I&E Quad Cup competition and secured dinner with faculty.

LiDAR3D ScanningEntrepreneurship

Duke I&E Competition Winner

Coursework

What I've studied

Computer Science

CS 372

Intro Applied Machine Learning

Applied machine learning course covering model development, evaluation, and practical ML workflows.

Mathematics

MATH 466

Math of Machine Learning

Mathematical foundations of machine learning, including linear algebra, optimization, probability, and model analysis.

Computer Science

CS 376

Computational Approaches to Human Language

Natural language processing course focused on computational methods for analyzing and modeling human language.

Computer Science

CS 370

AI/ML Applications

Machine learning algorithms, artificial intelligence concepts, and practical applications.

Computer Science

CS 330

Design and Analysis of Algorithms

Advanced algorithmic techniques including sorting, searching, dynamic programming, graph algorithms, and computational complexity analysis.

Computer Science

CS 316

Database Systems

Database design, SQL, transaction processing, and modern database management systems.

Statistics

STA 199

Introduction to Data Science

Statistical analysis, data visualization, and computational methods for extracting insights from data.

Statistics

STA 240L

Probability for Statistics

Mathematical foundations of probability theory with applications to statistical inference and data analysis.

Mathematics

MATH 216

Linear Algebra

Vector spaces, linear transformations, eigenvalues, and applications to computer science and statistics.

Mathematics

MATH 212

Multivariable Calculus

Calculus of several variables, including vectors, partial derivatives, multiple integrals, and applications.

Computer Science

CS 201

Data Structures & Algorithms

Advanced data structures, algorithm analysis, and problem-solving techniques for software development.

Computer Science

CS 250

Computer Architecture

Digital logic, processor design, memory systems, and assembly language programming.

Computer Science

CS 230

Discrete Math

Mathematical foundations for computer science including logic, sets, relations, functions, and proof techniques.

Psychology

PSY 490S

What Makes Us Human?

Special topics seminar in psychology exploring the uniqueness of human social learning and joint attention.

Language

SPAN 302

Spanish Grammar & Composition

Advanced grammatical structures, composition techniques, and linguistic analysis of the Spanish language.

Contact

Say hello

Email reaches me fastest. Happy to talk about ML, anything above, or something you're building.