Software engineer focused on AI-integrated system design — embedding LLM and computer vision capabilities into practical, end-to-end applications. I bring a product-minded approach to architecture decisions: designing modular systems, owning features from concept to implementation, and solving the real problems behind the requirements.
What I do
Embedding LLMs and ML models into real products — prompt pipelines, knowledge grounding, and practical inference workflows.
Multi-component architectures with clear module boundaries, inter-service communication, and decisions that hold up over time.
End-to-end ownership from database schema and REST APIs to polished React and React Native interfaces.
Clean architecture, design patterns, and reliable business logic — built to be maintainable, not just functional.
Selected projects
Designed the overall system architecture and led LLM integration via Gemini API — implementing a knowledge base construction layer using Java Reflection, defining module boundaries that enabled teammates to independently develop the UI and chat history components. The prompt pipeline combines user input, encoded domain knowledge, and prompt instructions, delivering ~1s responses with guardrails for out-of-scope queries. Business logic, knowledge encoding, and UI are fully decoupled.
Designed and owned the MVC architecture for a desktop app that continuously captures and analyzes student engagement during live sessions. Built the full View layer and Controller logic coordinating three independently developed modules — eye tracking, facial expression recognition (DeepFace), and Firebase database — into a single stable application. Diagnosed and resolved a cross-module camera resource conflict by centralizing instantiation in the DataRouter class, ensuring a single shared instance across all components within a session.
Independently designed and built the entire client-server system: defined the MongoDB schema, implemented custom user authentication, engineered detection history storage, and integrated Roboflow's pre-trained breed detection model via API. Made every architecture decision solo — from REST endpoint design to client-side state management. Delivers results in ~2s with breed name, sample images, and descriptions.
Built the course recommendation module within a schedule generation system. Modeled major and elective requirement rules as structured JSON based on official program requirements, and implemented a rule-matching engine to generate personalized course suggestions. Parsed student transcript data from XHTML to JSON, feeding structured input into the recommendation pipeline end-to-end.
Led development of a Java desktop app modeling a multi-enterprise production pipeline across four enterprises: Smartphone, Component Supplier, EMS, and Logistics. Designed inter-enterprise work request routing and ownership boundaries to reflect real-world delegation patterns from design through production and distribution.
Proposed and trained YOLOv8 for bird species classification after researching model alternatives, achieving 99% accuracy on a 2GB image dataset with GPU acceleration. Implemented model inference integration into a teammate-built Flask web application, enabling interactive species identification.
Implemented distributed Visual Question Answering training with FSDP and DDP in PyTorch, using CLIP for multimodal feature extraction and Transformer for cross-modal fusion. Configured and ran experiments on HPC clusters via Slurm with mixed-precision optimization.
Technical skills
Languages
Backend & Systems
AI & ML
Frontend & Tools
Experience
Penn State Behrend · Digital Humanities Lab · Erie, PA
Education
Northeastern University · Boston, MA
Penn State Erie, The Behrend College · Erie, PA
Open to AI application, backend, and full-stack engineering roles — available from Summer 2026.