Jarod Bloch

A live demo of hidden instructions that AI assistants pick up when they read a webpage, planted in page text, alt-text, metadata, and Unicode obfuscation. Triggered acknowledgment from Claude, ChatGPT, Gemini, and Copilot. This was the homepage; it now lives on its own subpage so it doesn't interfere with automated reads of this one.

A framework that maps how files in a codebase depend on each other so LLMs can annotate types across an entire repository, not just one file at a time. Reduced type errors by 37%, validated across Qwen3 models from 0.6B to 30B.

An autonomous greenhouse that grew lettuce and radishes for weeks with minimal human intervention. Fine-tuned vision pipeline (UNet, ~97% plant-pixel accuracy) running on a Raspberry Pi, with anomaly detection across 13 environmental sensors.

Yi Fei Cheng*, Jarod Bloch*, Alexander Wang, Andrea Bianchi, Anusha Withana, Anhong Guo, Laurie M. Heller, David Lindlbauer (*co-first authors)

Summary: Investigated how giving AI agents a physical presence using only spatialized audio and situated Foley sound effects changes the way we perceive and interact with them.

ML engineering and research roles: agentic systems, LLM pipelines, and infrastructure front and center.

Security-focused roles: adversarial ML, red teaming, prompt injection, and reverse engineering up top.

Safety and alignment roles: robustness evaluation, jailbreak research, and guardrails emphasized.