Hi, I’m Massad.

I ship AI agents that run in production, not just in notebooks.

At the Rutgers Turfgrass Pathology Lab I’m architecting a 33-node Python/LangGraph pipeline — a full-stack system (React/TypeScript, FastAPI, PostgreSQL) that orchestrates an 11-agent AI system and is designed to scale to 100+ concurrent users in production. The same lab is also where I built a hybrid disease-forecasting model, a viable alternative to the classic Smith-Kerns baseline on real held-out data.

See the pipelineResumeContact
INGESTRETRIEVE11 LLM AGENTSFAULT-CHECKQUALITY GATESSHIP
pipeline status — nominal — 33 nodes active

Orchestration

33-node LangGraph pipeline, 11-agent AI system across weather, pest, herbicide, and news domains.

Subscriber Platform

React/TypeScript + FastAPI + PostgreSQL, live in 6,800+ ZIP codes across the Northeast and Midwest.

Fault Tolerance

Automated retry logic plus plagiarism, cross-check, and citation-audit gates before anything ships.

The path here

Nov 2025 to now

Nov 2025

ML Research Assistant, Rutgers Turfgrass Pathology Lab

Built a two-stage hybrid ensemble (Random Forest + XGBoost/LightGBM) to outperform the Smith-Kerns baseline — 97% of real disease cases caught, 0.987 AUC on a held-out test season.

Jan 2026

StudyBuddy

Multi-agent tutoring system: 6 LangGraph agents over a pgvector semantic search pipeline, full-stack on React/TypeScript + FastAPI.

Mar 2026

AI/ML Fellow, Break Through Tech — Cornell Tech

Selected from 4,000+ applicants. Classified 13,200+ NYC Airbnb listings via a tuned Decision Tree / Neural Network comparison — 82.8% accuracy.

Mar 2026

Data Analytics & AI Skills Playbook

32-skill AI-assisted engineering library across 8 domains, cutting pre-ship review time 95% and closing a fact-checking gap in production.

Apr 2026

Software Engineer, Rutgers Turfgrass Pathology Lab

Automating the lab's weekly newsletter via a 33-node LangGraph pipeline and an 11-agent AI system, cutting research time 95% for superintendents across 6,800+ ZIP codes in the Northeast and Midwest.

Present

Manuscript in preparation

First-authoring the dollar spot forecasting results with Dr. Joseph Cohen and Dr. Ming-Yi Chou.

How the pipeline is built

A 33-node LangGraph system, end to end

1

Orchestrate the pipeline with LangGraph

A 33-node Python/LangGraph pipeline runs an 11-agent AI system (Anthropic + OpenAI APIs) across weather, pest, herbicide, and news domains, cutting 5–10 hours of manual research and writing per week.

2

Build the subscriber-facing surface

A React/TypeScript frontend talks to a FastAPI backend for the newsletter signup flow — an admin dashboard and automated SendGrid confirmation emails, backed by PostgreSQL, eligible for 6,800+ ZIP codes across the Northeast and Midwest.

3

Design the data layer

PostgreSQL-backed retrieval eliminates manual chemical handbook lookups, delivering zip-specific spray and fungicide guidance across weather, pest, herbicide, and news sources.

4

Make it fault-tolerant

Automated retry logic plus plagiarism, cross-check, and citation-audit gates catch bad output before it ships — reliability engineering, not just model accuracy.

5

Ship it and keep it running

Deployed with cron-scheduled production runs — the pipeline ships newsletters on its own schedule, unattended, and is designed to scale to 100+ concurrent users.

33node DAG

Projects

What I’ve been building

Apr 2026 — PresentRutgers School of Environmental and Biological Sciences

Turfgrass Weekly — A Newsletter Powered by AI Agents

  • Built and deployed an autonomous multi-agent newsletter pipeline that has run consecutive weekly production cycles, orchestrating 30+ LangGraph nodes across ingestion, writing, and quality-control layers with dynamic parallel fan-out per subscriber ZIP code.
  • Automatically caught and corrected unsupported factual claims in ~67% of AI-generated sections before publication by building a Claude-based fact-checking layer that cross-references every claim against its cited source and triggers a targeted revision when unsupported.
  • Eliminated LLM cost and latency for structured-data sections by routing time-sensitive pest/weed content around GPT entirely, rendering it directly from parsed agricultural extension model data and reserving GPT-4o calls only for sections requiring real synthesis.
  • Automatically screened and rewrote AI-generated content flagged for plagiarism in real time, achieving a 90% clean rate across production runs, by building a GPT-4o-mini plagiarism guard with an automatic rewrite-and-recheck loop before publication.
  • LangGraph
  • Claude
  • GPT-4o
  • Multi-Agent
View on GitHub ↗
Mar 2026 — Present

Data Analytics & AI Skills Playbook

  • Built and open-sourced a 32-skill AI agent library — reusable prompts for ML auditing, data-leakage/drift detection, and production-readiness reviews — portable across Claude Code, GitHub Copilot, and Codex, published as a live documented site.
  • Designed 6 multi-agent orchestration workflows in Claude Code that fan out parallel subagents per audit dimension and run independent adversarial verification on every finding before reporting, cutting false positives versus a single-pass review.
  • Validated the system against a production LangGraph multi-agent pipeline (9 concurrent data-ingestion agents, FastAPI + pgvector RAG): surfaced 19 real reliability/idempotency/security issues, all confirmed under independent re-verification.
  • Extended the toolkit with skills for staged ML models and LLM pipeline concerns (training/serving skew, RAG index health, LLM cost/routing audits), generalizing lessons from two production projects into reusable, cross-project review tooling.
  • Claude Code
  • AI Agent Skills
  • Open Source
View on GitHub ↗
Nov 2025 — Apr 2026Rutgers School of Environmental and Biological Sciences

Two-Stage Hybrid Ensemble Machine Learning for Dollar Spot Forecasting

  • Engineered a two-stage seasonal-switching forecasting pipeline (RandomForest gate → XGBoost/LightGBM regressors) that dynamically routes each prediction between a recursive early-season model (weather + foci-history past covariates) and a non-recursive late-season model (weather-only), with switch points driven by a Smith-Kerns logistic risk threshold (RH/AT coefficients, 5-day window, threshold=20.0) rather than a fixed calendar date.
  • Built a leakage-safe feature pipeline in shared_pipeline.py with lag/rolling covariates (3- and 7-day windows) across temperature, humidity, dewpoint, and rainfall, plus engineered disease-momentum features (foci_slope_lag1-3, Foci_growth_roll3/7, AT_range_roll3/7) informed by correlation analysis (e.g., slope_lag3 r=0.655 with foci slope) to capture disease-crash dynamics from cooling trends.
  • Achieved a Stage 1 outbreak-gate ROC-AUC of 0.987 and average precision of 0.979 on a held-out 2017 test year (grouped by cultivar susceptibility — Highly/Moderately/Resistant), with a tuned decision threshold (0.296) balancing an 88% precision / 97% recall operating point.
  • Ran a rigorous comparative evaluation against a baseline Smith-Kerns model across multiple outbreak-severity tiers (Mild/Moderate/Severe) and warning windows (1/3/5-day), using leave-one-year-out validation and confusion-matrix-derived precision/recall/F1/ROC-AUC to quantify the hybrid model's lift over the agronomic baseline.
  • A manuscript for this project is currently being drafted in collaboration with two Rutgers University professors from the Department of Mechanical & Aerospace Engineering and Department of Plant Biology.
  • RandomForest
  • XGBoost
  • LightGBM
  • Time Series
View on GitHub ↗
Jan 2026

StudyBuddy

  • Architected a multi-agent tutoring system with LangGraph, orchestrating 6 specialized LLM agents over a pgvector semantic search pipeline to deliver context-grounded Q&A.
  • Deployed a full-stack app (React, TypeScript, FastAPI) exposing 5 REST route groups across Vercel, Railway, and Supabase with JWT/OAuth auth, powering real-time mastery scoring across various topics.
  • LangGraph
  • pgvector
  • FastAPI
  • React
  • TypeScript
View on GitHub ↗
Dec 2025

Job Search Agent

  • Engineered an autonomous job search agent using n8n workflows, OpenAI GPT-4, and RapidAPI, processing 100+ daily job postings; implemented AI-powered match scoring and automated notifications, reducing manual search time by 90%.
  • Developed an AI job matching system leveraging GPT-4 API for NLP-based job description analysis; designed multi-step reasoning prompts with skill extraction capabilities, achieving 85%+ accuracy in candidate-job fit scoring.
  • Built a scalable data pipeline integrating RapidAPI with n8n workflows to aggregate multi-platform job data; architected ETL processes storing 3,000+ monthly listings with environment-based configuration and secure API key management.
  • Designed responsive HTML email templates with CSS styling and JavaScript template literals; integrated SMTP automation and Google Sheets OAuth2 API for persistent job tracking with title-based deduplication.
  • n8n
  • GPT-4
  • RapidAPI
  • ETL
View on GitHub ↗
Nov 2025

Music Data Mining Study

  • Analyzed 3,000+ song dataset in R to test popularity trends across decades using z-tests and chi-square tests, validating that post-2000 songs showed significantly higher popularity scores.
  • Built decision tree classifier using rpart to predict song mode from audio features (danceability, duration, year) with 70/30 train-test split and evaluated performance using confusion matrices.
  • Applied Bayesian reasoning to calculate posterior probabilities for song characteristics, computing likelihood ratios to update prior beliefs based on temporal music data from the 2010s.
  • R
  • Statistics
  • Decision Trees
  • Bayesian
View on GitHub ↗

Technical skills

Stack

Languages

  • Python
  • Java
  • JavaScript
  • TypeScript
  • SQL
  • HTML/CSS
  • R

Developer Tools

  • Git/GitHub
  • Docker
  • Jupyter Notebook
  • VS Code
  • n8n
  • RStudio

ML / Data Science

  • Scikit-Learn
  • XGBoost
  • LightGBM
  • Darts
  • Pandas
  • NumPy
  • Matplotlib
  • Seaborn
  • GridSearchCV

AI / LLM

  • OpenAI API
  • Anthropic API
  • LangChain
  • LangGraph
  • Prompt Engineering
  • Loop Engineering
  • Claude Code

Web / Cloud

  • FastAPI
  • RESTful API Design
  • TypeScript
  • Tailwind CSS
  • PostgreSQL/pgvector
  • SQLAlchemy
  • Vercel
  • Railway
  • Supabase

Right now

What I’m working on

Massad Raza

About

Why turfgrass pathology, of all things

I’m a Computer Science and Data Science student at Rutgers. Most of my classmates’ portfolio projects are recommendation engines or chatbots — mine is a disease-outbreak model for golf courses, because that’s the lab that took a chance on a sophomore who could code and wanted to work on something that mattered to a real person’s job. Turf managers make spray/no-spray calls that cost real money and affect real turf; getting them a few days’ more warning turned out to be a genuinely interesting systems problem, not just an ML exercise.

Outside of that, I work across the stack — Python for data/ML, TypeScript and React for interfaces, LangGraph/LangChain when a project needs an actual pipeline instead of a single model call. I like finishing things enough to put them in front of people, which is most of why this site works the way it does.

Rutgers University

B.S. Computer Science and Data Science — Expected May 2028

GPA 3.82  ·  Dean’s List Fall 2025, Spring 2026
CourseworkObject-Oriented Programming, Calculus, Data Structures & Algorithms, Discrete Mathematics, Linear Algebra, Probability & Statistics, Computer Architecture