Robert Oschler

I am a veteran software architect and systems developer with many years of experience engineering natural language processing, intent discernment, and data infrastructure pipelines. My background has been focused on adapting large language models into larger systems that include deterministic components, intended to improve the reliability and safety of A.I. agents I have extensive experience in hybrid RAG/LLM search utilities involving Hierarchical Navigable Small World (HNSW) combined with BM25 full-text indexing techniques, to create large-scale distributed backend databases that support semantic searches.   I also have skill using local LLM models like the Gemma 4 - e4b and the built-in Chrome browser LLM, Google Nano, with the end game being private RAG retrieval engines that execute entirely within the browser runtime using Transformer.js and WASM modules, thus facilitating semantic applications with a zero-egress browser architecture. My work has been recognized by industry bodies, including winning first place in the Story Protocol Buildathon for advanced semantic condition detection and being a member in the Pingcap Champions list for MySQL database application excellence. In the AI safety space, my focus is strictly technical: engineering runtime constraint layers, evaluating token-level logprobs, and building localized execution boundaries to prevent autonomous agent trajectory failures.

Personal Website

https://github.com/roschler

LinkedIn

LinkedIn

Projects

No grants recorded.