Graduated Summa Cum Laude

In progress

• Architected and deployed end-to-end AI-driven data pipelines for 8 clients, leveraging Python, PostgreSQL, AWS S3, and Supabase to process unstructured data into production-ready insights at scale.
• Designed a multimodal inference pipeline combining OCR and Vision-Language Models (LLaVA, CLIP) to extract metadata across 500,000+ items, facilitating $2M in inventory digitization and an $18K/mo revenue lift.
• Implemented LLM-based validation systems using local models (Ollama) and OpenAI APIs to score data quality and detect anomalies, reducing operational costs by $160K/year for healthcare and retail partners.
• Built asynchronous processing systems using Python (asyncio, multiprocessing) and queue-based workflows to handle high-throughput parallel inference workloads.
• Developed automated pricing engines integrating external web scraping, embedding-based similarity search, and heuristic ranking models to provide real-time market recommendations.
• Led a team of 4 engineers, establishing CI/CD pipelines (GitHub Actions), code review standards, and modular service architectures to support rapid deployment cycles.

• Designed and implemented a modular ROS 2–based robotics simulation stack for 6-DOF robotic manipulators using MoveIt 2, Gazebo, and RViz, enabling automated validation of motion planning and control algorithms.
• Developed motion planning pipelines integrating inverse kinematics, trajectory optimization, and real-time collision detection under dynamic constraints.
• Built reusable ROS nodes and services to support scalable simulation workflows, including planning, execution, and environment interaction.
• Integrated real-time perception systems (depth sensors, point clouds) to generate 3D occupancy grids and voxel-based representations for obstacle avoidance and spatial reasoning.
• Implemented sensor fusion techniques to combine multiple perception inputs, improving robustness in partially observable and noisy environments.
• Optimized simulation performance using GPU acceleration (CUDA) and efficient state update strategies, enabling large-scale stress testing and faster iteration cycles.
• Conducted systematic validation of planning algorithms through simulation-based testing, improving reliability prior to deployment on physical systems.

Internship

• Developed automated testing infrastructure for web applications using JavaScript/TypeScript and Selenium, enabling real-time validation of production deployments.
• Designed and implemented end-to-end testing pipelines integrated into CI/CD workflows (GitHub Actions, Jenkins) to detect regressions and enforce quality gates before release.
• Built data validation and monitoring scripts to track application performance, detect anomalies, and ensure consistency across staging and production environments.
• Created automated test suites covering UI, API, and integration layers, improving test coverage and reducing manual QA effort.
• Collaborated with cross-functional engineering teams to define testing standards, validation metrics, and release criteria for production systems.

DAIS Lab at the University of Houston

• Conducted research in machine learning, data systems, and computer vision, focusing on LLM forecasting, generative models, and graph-based query systems.
• Designed and evaluated deep learning architectures (LSTNet, Transformers) for time-series forecasting and structured prediction, leading to publication in ISPRS Annals (2025).
• Implemented transitive closure computations over graph-structured data (G+), developing recursive query pipelines and validating equivalence between SQL-based and in-memory approaches.
• Designed efficient data processing and experimentation pipelines in Python (Pandas, NumPy) for large-scale model evaluation and benchmarking.
• Explored diffusion-based generative models for planetary horizon reconstruction using DEM-guided constraints, contributing to novel research in computer vision and remote sensing.
• Translated research findings into reproducible systems and prototypes, bridging theoretical methods with practical data engineering implementations.

AI/ML Engineer specializing in LLM-driven validation systems, large-scale data pipelines, and decision-support tools. Proven ability to architect end-to-end multimodal systems, automate quality assessment, and lead teams to deliver measurable ROI.

Python, PostgreSQL, Ollama, Hugging Face, OCR, FAISS

• Architected a high-throughput multimodal data pipeline combining OCR (Tesseract, EasyOCR) and VLMs (BLIP-2) to extract structured metadata from noisy image and text inputs at scale.
• Designed data normalization and validation layers using LLMs (OpenAI APIs, Ollama) to enforce schema consistency, resolve ambiguous attributes, and standardize outputs across heterogeneous data sources.
• Implemented embedding-based similarity search using FAISS to match products, deduplicate entries, and infer missing attributes from historical datasets.
• Developed asynchronous processing workflows (Python multiprocessing/asyncio) to parallelize ingestion, inference, and post-processing for large-scale datasets.
• Built automated pricing and trend analysis modules integrating external web data, heuristic ranking, and similarity-based retrieval to generate real-time pricing recommendations.
• Benchmarked local inference (Ollama, GPU-based) vs. API-based models, optimizing for latency, cost efficiency, and output reliability across 500,000+ items.
• Designed end-to-end data flow architecture (ingestion → processing → validation → storage) using Python and PostgreSQL for scalable analytics and downstream integration.

AWS S3, Supabase, Python, Microsoft Graph API, Vercel, Render, PostgreSQL

• Engineered a distributed ingestion pipeline processing 10,000+ documents/day, utilizing AWS S3 for object storage and Supabase (PostgreSQL) for structured indexing and querying.
• Built automated ETL pipelines (Pandas, NumPy) to clean, normalize, and transform semi-structured clinical notes into analysis-ready datasets.
• Developed an LLM-powered validation system (OpenAI APIs, local models) to analyze clinical documentation, detect inconsistencies, and generate structured quality metrics for operational decision-making.
• Implemented event-driven and scheduled workflows to ensure reliable, near real-time data synchronization across ingestion, processing, and reporting layers.
• Automated PDF generation and secure document synchronization using Microsoft Graph API, enabling structured file distribution across organizational units.
• Designed RESTful API endpoints to expose aggregated metrics and processed data to downstream applications and dashboards.
• Deployed backend services on Render and frontend interfaces on Vercel, configuring environment management, API routing, and production builds for scalable cloud hosting.
• Eliminated manual QA workflows and administrative overhead, resulting in $160K/year cost savings and improved data visibility across clinical operations.

Python, React, Next.js, PostgreSQL, Web Scraping, Render, Vercel

• Architected and deployed a full-stack analytics platform to ingest and visualize clinical documentation from third-party EHR systems lacking public APIs, utilizing authenticated session-based scraping pipelines.
• Designed ETL workflows in Python (Pandas, NumPy) to normalize raw note data and built automated ingestion pipelines with scheduled background workers for near real-time synchronization across thousands of records.
• Engineered a modern frontend using React and Next.js with dynamic data visualizations (Recharts/D3.js) and a REST API backend to track clinician productivity, note completion rates, and service delivery trends.
• Integrated PostgreSQL/Supabase for scalable historical trend analysis and deployed the infrastructure on Render and Vercel, enabling leadership to reduce manual reporting overhead through automated, data-driven insights.

Next.js, React, Vercel, Supabase, PayPal API, OpenStreetMap, Google Cloud

• Designed and developed a full-stack production website from scratch using Next.js and React, implementing server-side rendering and dynamic routing for performance and SEO optimization.
• Integrated secure payment processing using PayPal APIs to support memberships and donations, enabling automated transaction handling and user onboarding.
• Built and managed a Supabase (PostgreSQL) backend to track membership data, user records, and payment status, supporting real-time updates and administrative workflows.
• Implemented interactive mapping features using OpenStreetMap and Google Maps APIs to display event locations and points of interest, enhancing user engagement and accessibility.
• Developed custom frontend components and responsive UI layouts to support content management, navigation, and user interaction across devices.
• Deployed and maintained the application on Vercel, configuring environment variables, domain routing, and production builds for reliable hosting.
• Collaborated with stakeholders to translate organizational needs into technical features, delivering a scalable platform for community engagement and operational management.

Diffusion Models, ControlNet, LoRA, Computer Vision, Geospatial Data, PyTorch, TensorFlow

• Developed a novel geospatial representation system (DESPINA) to synthesize geometry-preserving planetary horizon imagery from digital elevation models (DEMs), addressing the scarcity of ground-level data.
• Designed a multimodal pipeline combining DEM-derived structural embeddings (inverse-depth maps, soft-edge constraints) with text conditioning to guide diffusion-based image generation.
• Integrated Stable Diffusion (SDXL) with ControlNet and custom LoRA fine-tuning to enforce geometric consistency while generating visually realistic terrain reconstructions.
• Implemented a custom DEM-to-depth pipeline using ray-casting and inverse-depth normalization to preserve skyline fidelity and terrain structure across varying viewpoints.
• Generated structured training datasets by combining Apollo imagery, depth maps (DPT), soft-edge features (PidNet), and captioning models (BLIP) for supervised conditioning.
• Designed and executed ablation studies evaluating the contribution of depth, edge, and text constraints, demonstrating improved structural similarity and horizon accuracy over image-conditioned baselines.
• Established a geometry-first evaluation framework using DEM-derived skylines and terrain masks to quantitatively validate reconstruction fidelity independent of photographic data.
• Demonstrated domain-agnostic applicability across planetary bodies (Moon, Earth, Mars), enabling scalable generation of synthetic datasets for downstream tasks such as visual place recognition.