Huifeng Jason Li

Title: Huifeng Jason Li and Hybrid Retrieval for Generative AI.

Introduction

Huifeng Jason Li is an inventor based in Highland Park, New Jersey. His patent work addresses information retrieval for retrieval-augmented generation (RAG), a framework that supplies large language models with relevant source material when generating answers.

Latest Patents

Li is associated with one patent, titled “Systems and Methods for Hybrid Lexical-Vector Retrieval in Retrieval-Augmented Generation Models.” The invention combines lexical search, which finds text based on matching terms, with semantic vector search, which identifies content by meaning. It preprocesses user queries using techniques such as named-entity recognition and date standardization, and may also detect queries that fall outside the system’s domain.

The system merges and ranks results from both search methods, prioritizing overlapping results and interleaving other candidates according to a predetermined rule. It then selects relevant document chunks for a large language model, along with instructions tailored to the query’s detected intent. The model uses the retrieved material to produce a grounded answer in a standardized format.

Career Highlights

Li’s listed patent work focuses on improving how RAG systems retrieve and organize information before generating responses. He is associated with Morgan Stanley Services Group Inc. The available information does not specify further details about his role or career history.

Collaborations

Jing Shen is identified as Li’s coworker. The available information does not establish whether Shen collaborated on the patent.

Conclusion

Huifeng Jason Li’s patent describes a hybrid retrieval approach that combines keyword matching and semantic search to provide large language models with relevant, intent-specific source material.