Honeoye Falls, New York, United States
Hi! I’m Robert Hochgraf - Research Engineer @ Re:Build Manufacturing, Music Producer, and Aurora Borealis Watcher. You can find me around Rochester, NY - ➛ Dreaming about Post-Training Techniques ➛ Researching High Performance Computing @ University of Rochester ➛ Composing Music for a Student-Led Film ➛ Developing the Indirect App [https://www.instagram.com/indirect_app/] I love meeting new people - please feel free to DM me with project/research ideas or general friendliness! You can also get in touch with me via my website: https://rhochgraf21.github.io/cv2
➛ Optimized local LLM inference using vLLM, leading to 5x improvements in token speed generation. ➛ Prototyped proof-of-concept RAG pipeline using vector retrieval to enable demo of end-to-end retrieval performance to stakeholders. ➛ Researched and deployed MCP servers to enable agents to perform actions such as web search.
➛ Developed chatbot interface that automatically transforms company data into business knowledge. ➛ Contributed to data retrieval systems, including by authoring PRs adding agentic planning, hybrid graph-vector search, and RAG optimizations (eg, reranking). ➛ Authored prompt engineering copilot to streamline prompt engineering process while also improving prompt performance by up to 50%.
➛ Formulated novel performance tuning framework to find GPU kernels that are portable and high performance. ➛ Achieved up to 7x speedups over a state-of-the-art linear algebra library with potential to impact open-source machine learning projects such as llama.cpp and volunteer scientific computing projects. ➛ Presented findings at Finger Lakes High Performance Computing Symposium & University of Rochester Systems seminar.
➛ Explored and documented potential for machine-learning enabled capabilities. ➛ Researched, modified, and documented test-range infrastructure to support deployments onto production. ➛ Automated onboarding processes for team members.
Spearheaded the development of an industry-first education mobile app to enhance the hand-eye coordination of dental professionals. Partnered with a seasoned dentist to pinpoint requirements and bring the app from ideation to market launch. Released the app on both Android & iOS to reach an addressable audience of 18,000+ dental and dental hygienist students. Ranked #25 in Education on iOS (as of Jan 2024).