Operations Research, AI/ML, and Data Science professional with experience across industrial operations, rail transportation, manufacturing, logistics, and applied analytics. I build optimization models (LP/MIP, vehicle routing), predictive systems, anomaly-detection pipelines, and decision-support tools using Python, SQL, Gurobi, CPLEX, SCIP, OR-Tools, machine learning, cloud platforms, and Docker.

Currently a Georgia Tech PIN Fellow working on reinforcement learning, HMM-based optimization, multi-objective modeling, and RAG applications. Previously supported operational analytics and railcar asset management at Norfolk Southern and conducted machine-learning research with Georgia Southern University and MxV Rail.

🔬  PIN Fellow @ Georgia Tech              →  HMM + RL optimization pipelines for manufacturing · RAG-based knowledge systems (2025–Present)
🚂  Supervisor Associate (Analytics and reporting) — Operations Division, Mechanical Maintenance @ Norfolk Southern → Rail equipment management, FRA/49 CFR compliance, real-time asset health monitoring, and operation optimization (2024-2025)
🎓  Research Assistant @ Georgia Southern  →  95–97% anomaly detection on live rail DAS datasets · AAR/TTCI collaboration (2022–2023)
📄  7 peer-reviewed papers (ASME, Springer, SPIE, Elsevier) — 200+ citations
🎯  Georgia Tech OMSCS (CS) →  Fall 2026 - Current

ai-last-mile-delivery-optimization  

End-to-end vehicle routing platform combining exact and heuristic optimization with ML forecasting. Formulates a Capacitated Vehicle Routing Problem with Time Windows (CVRPTW) two ways: an exact mixed-integer program in Gurobi for provable near-optimality on small instances, and a production-scale solver in Google OR-Tools (guided local search) that solves the full instance in seconds — the same exact-vs-heuristic trade-off used in large-scale commercial routing systems. Layered with an XGBoost demand forecast (14-day horizon, ~5% MAPE) and a Random Forest travel-time model that captures rush-hour congestion (R²=0.97) instead of assuming flat driving speed. Interactive Streamlit dashboard with live route maps (Folium), KPI tracking, and fleet-size scenario/sensitivity analysis. Full pytest coverage validating capacity and time-window feasibility. Python Gurobi OR-Tools Linear/Mixed-Integer Programming XGBoost Random Forest Streamlit Folium Vehicle Routing Scenario Analysis

llm-finetuning-engineering-domain

Two complementary fine-tuning pipelines on railroad AI and manufacturing domain data. BERT/RoBERTa classification: fine-tuned bert-base-uncased → 94.2% accuracy; roberta-base → 95.8% on 4-class DAS signal conditions. LoRA generation: Mistral-7B instruction-tuned with only 4.2M trainable params (0.06% of model) using QLoRA 4-bit quantization — ROUGE-L 0.68. Both models published on HuggingFace Hub →
Python BERT RoBERTa Mistral-7B PEFT LoRA QLoRA Hugging Face Transformers NLP Jupyter Notebook

engineering-knowledge-rag  

Production RAG pipeline grounded in 7 peer-reviewed publications (200+ citations). Retrieves domain knowledge via FAISS + SentenceTransformers (all-MiniLM-L6-v2, 384-dim cosine search), generates citation-backed answers with Flan-T5 — zero hallucination on domain specifics. Live on HuggingFace Spaces (Docker). Supports drop-in PDF ingestion to extend the knowledge base to any domain.
Python RAG FAISS SentenceTransformers Flan-T5 LangChain Streamlit Docker HuggingFace Spaces

railroad-anomaly-detection-cnn-lstm

Hybrid CNN-LSTM with sliding window for railroad condition monitoring via distributed acoustic sensing (DAS). Achieved 97% train position detection rate on live HTL fiber-optic datasets from AAR/TTCI, Pueblo CO. Published in Green Energy & Intelligent Transportation, Elsevier 2024.
Python TensorFlow CNN LSTM Time-Series Anomaly Detection Signal Processing

das-railroad-gru-lstm

GRU and LSTM models for train presence detection along fiber-optic DAS-instrumented track. 94% detection rate. Published in SPIE Journal of Applied Remote Sensing, 2024.
Python Deep Learning GRU Distributed Acoustic Sensing Rail Safety

train-scheduling-system

Java console-based Train Movement & Scheduling System with CRUD operations, station management, scheduling, file I/O, and full OOP architecture.
Java OOP File I/O Scheduling Algorithms

warehouse-visual-intelligence

Production multi-agent AI system for real-time warehouse monitoring, safety violation detection, layout optimization, and cost reduction. A 5-agent pipeline (Vision → Layout → Anomaly → Cost → Orchestrator) processes images via YOLOv8 + GCP Vision API and outputs $/day cost impact per detected inefficiency. FastAPI backend, Streamlit dashboard, GitHub Actions CI/CD.
Python YOLOv8 CrewAI LangChain Google Cloud Vision GCS FastAPI Streamlit Multi-Agent Computer Vision

cv-manufacturing-defect-detection

Real-time surface defect detection for steel manufacturing using YOLOv8 on the NEU Surface Defect benchmark (1,800 images, 6 defect classes). Achieves 95.2% mAP@50 with 2.1ms GPU inference. Exported to Intel OpenVINO IR format for 2–4× CPU speedup on Intel hardware. Extends published WAAM research (Georgia Tech / Springer 2026). Includes Colab training notebook and Streamlit demo.
Python YOLOv8 Intel OpenVINO Jupyter Notebook Streamlit Computer Vision Manufacturing QC Deep Learning

waam-hmm-rl-optimizer

Hidden Markov Model + Reinforcement Learning pipeline for material design optimization in Wire Arc Additive Manufacturing (WAAM). Deployed under Georgia-AIM grant at Georgia Tech. 5% improvement in material utilization. Peer-reviewed Springer publication (2026).
Python Reinforcement Learning HMM Manufacturing AI MLOps

[ai-polymer-optimisation-mip-v3] -coming soon

AI-guided optimization of MIP/CIP polymer film synthesis using physics-informed data generation, ML surrogates, and multi-objective Pareto optimization — without wet-lab experiments. Achieves 95.2% capture efficiency and 1.998 µm thickness (±0.002 µm of target). Target: JOM / Springer (2026).
Python scikit-learn Random Forest MLP Gaussian Process Latin Hypercube Pareto Optimization Materials AI

manufacturing-powerbi-demo

Power BI + T-SQL demo for manufacturing analytics and real-time monitoring dashboards.
T-SQL Power BI Manufacturing Analytics

Building-a-Rainfall-Prediction-Classifier

End-to-end ML classification pipeline for rainfall prediction using supervised learning, feature engineering, and model evaluation.
Python scikit-learn Jupyter Notebook Classification

Store-Recommendation-System-Atlanta-GA

Interactive store recommendation system with analytics, K-Means clustering, and Folium map visualizations for Atlanta, GA.
Python Clustering Geospatial Recommender Systems Jupyter Notebook

Python-and-ML-Learning-Apps

Interactive Python & Machine Learning course with 28 structured lessons, built with React.
JavaScript React EdTech Machine Learning

📚 Full publication list on Google Scholar →  |  200+ total citations

• 🚚 Last-Mile Delivery Optimization — Extending the CVRPTW model with stochastic demand and split-delivery formulations
• 🏭 WAAM AI Prototype — Deploying HMM + RL material design system at Georgia Tech (Georgia-AIM)
• 🏗️ Warehouse Visual Intelligence — Extending with real-time RTSP camera feed + BigQuery analytics trend dashboard
• 🔬 CV Defect Detection — Fine-tuning PPE detection model; Vertex AI deployment pipeline
• 🤖 LLM Fine-tuning — Scaling LoRA Mistral-7B training dataset; evaluating RAG vs fine-tuned generation

I'm actively seeking roles in Operations Research, ML Engineering, AI/Data Science, Data Engineering, and AI Research — particularly in manufacturing, logistics, transportation, railway, energy, infrastructure intelligence, or large-scale ML and optimization systems.

Open to: Full-time roles at tech & industrial AI companies.