About
I am a Data Scientist and Engineer with a research focus on geometric representation learning for perception under real-world constraints. My approach to AI is defined by a "full-stack" philosophy: I believe that reliable perception systems require both theoretically sound representations and robust data infrastructure.
My core technical foundation is 3D perception and geometry-aware deep learning. At Oak Ridge National Laboratory (ORNL), I work with real sensor data (LiDAR) where noise, ambiguity, and class imbalance are the norm. My interests center on how meaningful geometric structure in learned representations affects robustness and uncertainty. I’ve seen that when representations fail to preserve underlying structure, models become brittle on novel configurations, making reliability a representation problem, not just a modeling one.
I tackle these research challenges with the discipline of a production engineer. My background includes over 5 years of experience in Data Analytics in companies like The AES Corporation, and Executel, where I architected semantic layers in BigQuery, optimized SQL pipelines for SAP data, and implemented CI/CD workflows. I understand that state-of-the-art models are useless without a robust data foundation, and I bring this rigorous, systems-level thinking to my research code and evaluation pipelines.
I am particularly interested in building perception systems that can serve as reliable foundations for embodied intelligence and robotic manipulation. I leverage my dual expertise, bridging the gap between raw data infrastructure (SQL, ETL) and advanced perception models (PyTorch, Geometric DL), to support data-efficient learning and safe downstream decision-making.
Self-Validating Agentic System for Bank Statements Data Extraction
OpenAI API • Agentic AI Workflow • OCR + text parsing
- Designed and implemented an agentic AI system to extract structured financial data from heterogeneous bank statement PDFs (TEXT and scanned VISION formats).
- Built autonomous routing logic to detect PDF modality and trigger specialized extraction pipelines.
- Implemented multi-agent validation with evidence-grounded PASS / FAIL / UNCERTAIN checks and bounded self-repair loops.
Geometric Deep Learning-Based Plant Organ Segmentation and Phenotyping Using LiDAR Point Clouds
Point cloud segmentation • geometric deep learning • robust evaluation
- Developed a geometry-aware deep learning pipeline for plant organ segmentation from 3D LiDAR point clouds.
- Engineered local geometric descriptors and evaluated under class imbalance and partial observability (occlusion).
- Current best configuration reports ~65.6% mIoU and ~82% stem recall.
EEG Motor Imagery Classification (Conv1D CNN)
Signal processing • deep learning • generalization challenges
- Built a full pipeline to classify left vs. right fist motor imagery from PhysioNet EEG (20 subjects, 64 channels, 160 Hz).
- Compared CSP-based baselines vs deep models; final 2-layer Conv1D CNN reached 76.61% accuracy (ROC-AUC 0.79).
- Analyzed subject-wise generalization (avg. accuracy 51.8%) and documented future work for domain adaptation and transfer learning.
CRM → PostgreSQL ETL for Real Estate Analytics (Render + Power BI)
Data engineering • event-driven ingestion • BI system design
- Built a cloud BI system that streams CRM events via Flask webhooks into PostgreSQL on Render.
- Combined real-time ingestion with scheduled incremental API sync, preserving full event history for auditability.
- Delivered 8 interactive Power BI dashboards using a standardized DAX measure library (60+ measures).
Predicting Hospital Readmission (Logistic Regression in R)
Statistical modeling • interpretability • healthcare analytics
- Modeled hospital readmission risk on 25,000 patient records using logistic regression with stepwise AIC selection.
- Achieved ~61.7% test accuracy (AUC 0.662) and identified key drivers (length of stay, prior visits, medication/labs, age).
Apex Capital: Financial + ESG Decision Support System (Power BI + Python + MySQL)
ETL + data integrity • decision support • analytics engineering
- Built a full-stack prototype DSS for “dual-mandate” investing (value + ESG), merging multi-source financial and ESG datasets.
- Implemented an ETL workflow with data quality discovery (e.g., mismatched tickers, missing ESG coverage) and schema-enforced loading into MySQL.
- Delivered a Power BI “financial terminal” dashboard with context-aware DAX measures (sector benchmarking, investable gating, ESG null handling).
PDF Charge Parser (Python GUI)
Automation tool • PDF parsing • Excel reporting
- Built a lightweight desktop tool to parse telecom billing PDFs and extract roaming/long-distance charges into a clean Excel report.
- Implemented robust parsing with PyMuPDF and rule-based extraction, wrapped in a simple Tkinter GUI with progress tracking.
Goals
- Research direction: grow from geometry-aware 3D perception into embodied AI (perception → learning → action), with an emphasis on robustness and data efficiency.
- Near-term: turn the ORNL 3D LiDAR project into a formal research write-up and submission.
- Skill ramp: deepen PyTorch and modern CV tooling toward robotics-relevant perception (3D, multimodal fusion, self-supervised learning).
- Career: pursue roles and/or a PhD path aligned with perception for robotics and real-world intelligent systems.
References
-
Robert M. Price, Ph.D.
Professor of Applied Data Science, Graduate Coordinator
East Tennessee State University
pricejr@etsu.edu
Research Supervisor -
Jeff R. Knisley, Ph.D.
Professor of Applied Data Science
East Tennessee State University
knisleyj@etsu.edu
Research Supervisor -
Joseph Antone
Senior Project Manager (Retired)
The AES Corporation
Former Project Manager
josephantone38@gmail.com -
Caleb Bennett
Director, International Enrollment & Services
East Tennessee State University
Direct Supervisor — Graduate Assistantship
bennettcb@etsu.edu -
Iliana Berganza
Customer Service Team Leader
The AES Corporation
Former Co-Worker
Iliana.berganza@aes.com
Updates
- 2026 Searching for a PhD to make a transition toward embodied AI: robust 3D perception → learning for manipulation/HRI.
- 2026-01 Preparing 3D LiDAR Plant Organ Segmentation project write-up for submission.
- 2025-12 Completed M.S. in Applied Data Science at East Tennessee State University (GPA: 3.79).
- 2025-12 Successfully defended research internship presentation at Oak Ridge National Laboratory.