Brayden Abo

Sunnyvale, CA

• Develop batching and route-optimization services in a Java microservices platform processing 3.5M orders/day
• Extended Last Mile Delivery Optimization solver to handle new delivery constraints, reducing order cancellations and missing delivery items, avoiding $23M in GMV loss.
• Designed LMD batch assignment algorithms to support Vision Center orders with unique fulfillment constraints, to integrate a new delivery vertical and enable <3 hour lens delivery.

Hoboken, NJ

• Improved meal ingredient recognition F1-score by 9% by building a local VLM pipeline (Qwen2.5-VL, Llama3.2) enabling zero-shot deconstruction of complex, multi-layered food items for Plato, a multimodal nutrition chatbot.
• Designed a multimodal mapping system that decomposes meal images into atomic ingredients and groups them under standardized food entities, enabling structured nutritional analysis at scale.
• Architected a full-stack web application to deploy Plato as a research tool for external institutions, handling multimodal input (image + text), user session management, and nutritional data output.

Basking Ridge, NJ

• Developed a GraphRAG pipeline to enhance LLM query performance on Verizon's network database, achieving a 70-80% reduction in query time.
• Integrated Model Context Protocol (MCP) into an AI-agent framework, enabling standardized, secure access for agents to diverse data sources and tools.
• Built an AI-agent framework on a structured knowledge graph with Python and Llama3 to help network engineers diagnose topology intermittencies and speed up network decommissioning projects