Aaron Loera
AI Researcher | AI/ML Engineer
Computer Science graduate seeking entry-level opportunities in ML engineering, AI engineering, or data science where research depth and engineering execution can drive measurable outcomes.
Computer Science graduate seeking entry-level opportunities in ML engineering, AI engineering, or data science where research depth and engineering execution can drive measurable outcomes.
I'm a dedicated computer science graduate from the University of Texas at Tyler with a profound passion for artificial intelligence, machine learning, and data science. My work focuses on developing innovative AI solutions that can make a real-world impact, particularly in healthcare applications and security-based technologies.
Through various projects and my time as a deep learning research assistant, I've gained extensive experience in designing, training, and deploying AI models. From CNNs and Transformers for medical signal processing to Autoencoders and Isolation Forests for anomaly detection, I've developed a well-rounded skillset aimed at creating AI-centric solutions.
The University of Texas at Tyler
RideAlong | Tyler, TX
The University of Texas at Tyler | Tyler, TX
Data Annotation Tech | Remote
Texas Roadhouse | Tyler, TX
InsiderGuard AI is a production-oriented insider threat detection system serving multi-user analyst workflows via a Streamlit dashboard with Firebase authentication. It leverages User and Entity Behavioral Analytics (UEBA) through a hybrid anomaly detection pipeline (Autoencoder + Isolation Forest) to score and flag behavioral risk in real time across 125,000+ telemetry logs, supporting real-time inference on live behavioral telemetry streams and surfacing results as prioritized alerts.
This project pairs an XGBoost match predictor, tuned via Optuna hyperparameter optimization, with a 10,000-run Monte Carlo bracket simulator generating win probabilities for all 48 World Cup 2026 teams. The backend is a FastAPI REST API with Pydantic v2 schema validation, containerized in Docker and deployed to Render with a GitHub Actions nightly retraining pipeline, with live tournament odds and match probabilities surfaced through a public Streamlit dashboard.
This project fine-tunes a Vision Transformer to classify QR codes as legitimate or malicious, addressing the growing security risks posed by visually indistinguishable phishing codes. Beyond detection, the model's predictions are made interpretable through attention rollout, highlighting the global visual patterns the transformer relies on when identifying malicious intent. Overall, the work demonstrates the feasibility of image-only QR code security as a proactive defense against QR-based attacks.
This project focuses on detection of cardiovascular diseases from 12-lead ECG signals using a custom deep learning CNN tailored to capture cross-lead cardiac patterns. Beyond classification, the learned feature space is explored using a probing classifier to evaluate how well metadata (age, sex, etc.) is encoded in the model's representations. The work aims to enhance automated cardiovascular disease detection and support more reliable decision-making in healthcare settings.
This project leverages the power of Langchain, Ollama, and Chroma to create a document-based question answering system. It uses a vector database to store document chunks and retrieves relevant information based on user queries. The system is capable of ingesting CSV files, splitting them into manageable chunks, and embedding them into a vector space for efficient search and retrieval. The assistant, acting as a college advisor, answers questions by leveraging the stored data.
This project implements sidewalk lane detection using a fine-tuned YOLOv8 model to support safe navigation for visually impaired users in real-world environments. To prioritize user safety, detections are filtered with a high-confidence threshold, while additional classes such as bicycles and automobiles are incorporated to reflect realistic walking conditions. Overall, the system demonstrates how real-time, vision-based AI can improve pedestrian safety by providing reliable environmental awareness in everyday navigation.
Support for academic and applied AI research projects, including literature review, methodology design, and implementation of cutting-edge algorithms.
Rapid development of proof-of-concept machine learning solutions to validate ideas and demonstrate feasibility for various applications.
Custom deep learning models for real-world applications, from computer vision to natural language processing and signal analysis.