Experience

• Contributing to the architecture of a privacy-first urban sensing network CityOS, under Dr. Jorge Ortiz.
• Specifically engineering the Single-Locality Aggregation API, a system designed to ingest and process continuous data streams from fixed, localized environments.
• Developing a multi-modal sensing pipeline for pedestrian behavior analysis, integrating sensor data from cameras, IMUs to identify pedestrians and their activities.

• Architected a PostgreSQL schema and Python ETL workflow to streamline multimodal gait analysis, slashing data synchronization latency by 90% across a corpus of 300+ recordings.
• Developed a dual-stream gait recognition framework leveraging Transformer backbones to correlate video and accelerometer data, delivering a predictive performance of 0.98 AUC-ROC.

Delivered lectures, designed lesson plans, graded assignments, and provided personalized feedback for 50+ undergraduates.

• Deployed an end-to-end Vision Transformer (ViT) pipeline on PyTorch to directly parse multilingual academic records without traditional OCR dependencies.
• Optimized document throughput for an Italian fintech client, achieving 88% extraction accuracy and slashing processing latency by 90%.
• Part of the Pi School of AI Fall 2022 cohort.

• Built KQL pipelines to extract and clean 50,000+ Sentinel security alerts, creating a Pandas preprocessed dataset in PostgreSQL
• Trained an ensemble alert-triage classifier with Scikit-Learn (Random Forest + Gradient Boosting) on structured log features, reducing false-positive rates by 18% across 3 client SOC environments
• Fine-tuned RoBERTa-base using PyTorch and HuggingFace on 40,000+ raw alert descriptions to better capture severity, boosting F1 by 12% over the tabular-only model
• Containerized the end-to-end inference pipeline with Docker and deployed it as a FastAPI microservice on Azure ML, that cut average analyst triage time by 5 minutes per incident

• Engineered high-fidelity ROS simulations to validate autonomous navigation, deploying SLAM for instantaneous mapping and localization.
• Implemented robust traffic perception systems using YOLO and R-CNN in TensorFlow, securing a 93% detection accuracy.