Gerardo Montemayor

B.S. in Computer Science, Minor in Electrical & Computer Engineering | GPA: 3.5/4.0

Relevant Courses: Generative AI, Deep Learning, Machine Learning, Algorithms, Data Science, OOP & Data Structures

Seam AI | San Francisco, CA

• Built a conversational Sales Persona generation agent with FastAPI backend, evaluation and monitoring on Langsmith, and Typescript/React frontend, reducing customer time-to-search by 1 hour per Persona
• Trained a NN classifier using feature-based transfer learning to align research queries to retrievers, maintaining accuracy parity with Gemini2.0-Flash router and reducing p50 latency from 1.5s to 50ms
• Implemented a string utilities module for broad use in the application unifying logging, error handling, and normalization. Normalization reduced our job title FAISS index disk space by 9MB without impacting accuracy

AI Innovation Lab | IT@Cornell

• Developer for AI initiative at Cornell building GenAI-powered apps to support campus organizations
• Building application integrating OCR technologies, VLMs, and LLMs to automate expense processing, ensure compliance with state/school policies, and reduce manual workload for financial team

Technology Advancement Center | Columbia, MD

• Developed a virtual assistant to support business operations, including front-desk interactions and company events
• Deployed a secure web application with Docker and AWS EC2, establishing a CI/CD pipeline to streamline delivery of weekly MVPs and improvements based on user feedback

Cornell Mars Rover

• Programmer for project team that builds semi-autonomous rover to compete in University Rover Challenge (URC)

CS 6784

• Co-authored research paper on LuKA, our proposed KV cache design that compresses long contexts for more efficient LLM inference while preserving information
• Implemented sparse attention mechanisms and developed the evaluation framework for benchmarking performance, achieving 27% accuracy improvement over SOTA methods on multi-topic QA benchmarks with LuKA

Cornell Mars Rover

• Implemented a ROS2 package with OpenCV to process camera input to detecting tags in field, determine relative positions and orientations of tags, and navigate difficult terrain for Autonomous Challenge at URC

Cornell Mars Rover

• Implemented a high-performance interface to operate and access rover capabilities, including camera feeds, life-detection systems, robotic arm simulations, and a central ROS2 node launch hub
• Created specialized GUI components using React and Foxglove Studio, a robot visualization platform, enhancing rover control through seamless integration with ROS2

• Developed a Cornell-themed tower defense game in OCaml with player upgrades and bullet-tracking physics