Hi, I'm Jordan



You can find some of my projects listed here. I'm interested in discovering subtle insights in data and applying them to real-world problems.



Purpose

To develop and benchmark data science solutions for real-world problems.

Solution

Built ensemble model with unsupervised feature discovery for heart disease prediction. Competing in the Deep Past Challenge for the translation of ancient Akkadian trade documents.

Tools

PythonJupyterPyTorchTransformersFeature EngineeringHyperparameter OptimizationUnsupervised Learning

Impact

Top 6% in heart disease prediction (0.9553 AUC).

Purpose

To provide bus riders with 24/7 customer service that is easily accessible and real-time, across multiple languages.

Solution

Built AskLynchburg - a bilingual assistant that gives real-time bus times, nearby restaurants, and answers questions using real-time GTFS data integration and natural language processing.

Tools

PythonNLTKModel TrainingData PreprocessingFlaskPyTorchAWSREST APIs

Impact

Acquired by Systrans (EQUANS), scaled from local to international deployment

Purpose

Train reinforcement learning agents to make autonomously optimized decisions in simulated environments.

Solution

Implemented 12+ RL algorithms including Deep Q-Networks, Proximal Policy Optimization, Actor-Critic methods, and policy gradient with custom neural network architectures and reward optimization.

Tools

PythonPyTorchStable Baselines3GymnasiumNeural NetworksTensorFlow

Impact

Top 1% on Huggingface for the LunarLander-v2 project

Purpose

To provide GLTC with an efficient method of addressing common customer service questions/requests during peak business hours.

Solution

Built AskGLTC - a chatbot that answers rider questions, sends SMS updates, responds via SMS, and integrates with GLTC's WordPress site.

Tools

PythonFlaskWordPressTwilioMySQLAWS

Impact

Served 300+ riders daily, recognized by the Virginia DRPT, won $5,000 APTF scholarship



Skills


  • Python

    Primary language for ML, data pipelines, and web development. Used across most projects from NLP chatbots to reinforcement learning agents.


  • PyTorch

    Deep learning framework for NLP and neural network development. Used to build the intent classification models behind AskGLTC and AskLynchburg.


  • TensorFlow

    Framework for training reinforcement learning agents. Achieved top 1% on the HuggingFace Lunar Lander benchmark out of 7,000 submissions.


  • AWS

    Cloud infrastructure for production deployments. Deployed and monitored AWS resources supporting real-time transit systems for 11+ agencies nationwide.


  • SQL

    Database querying and management. Used for structuring and retrieving data across transit and geospatial data pipelines.


  • ArcGIS Pro

    Spatial analysis and geospatial data processing. Used to integrate GTFS route data for transit agencies across the U.S.


2026 · Jordan Romano · Built with React.js