Less Context, Same Performance: A RAG Framework for Resource-Efficient LLM-Based Clinical NLP
Satya Narayana Cheetirala 1 , Ganesh Raut 1 , Dhavalkumar Patel 1 , Fabio Sanatana 1 , Robert Freeman 1 , Matthew Levin 2,4 , Girish N. Nadkarni 3,4 , Omar Dawkins 6 , Reba Miller 5 , Randolph M. Steinhagen 7 , Eyal Klang 3,4 , Prem Timsina 1 · 2025
Read Less Context, Same Performance: A RAG Framework for Resource-Efficient LLM-Based Clinical NLP by Satya Narayana Cheetirala 1 , Ganesh Raut 1 , Dhavalkumar Patel 1 , Fabio Sanatana 1 , Robert Freeman 1 , Matthew Levin 2,4 , Girish N. Nadkarni 3,4 , Omar Dawkins 6 , Reba Miller 5 , Randolph M. Steinhagen 7 , Eyal Klang 3,4 , Prem Timsina 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-20320.txt.
Description
Satya Narayana Cheetirala and co-authors show that a language model needs only the best bits of a long clinical note. Their 2025 paper tackles a cost problem: full notes are long, and feeding them whole is slow and dear. The team asks whether retrieval can match whole-text scores.
Notes were split into chunks, indexed, and the top 4,000 words for each query were passed to the model. Three models were tested on spotting surgical complications. Scores for the retrieval route matched whole-text runs, with no real gaps on any metric. The authors pitch the method as a cheaper, scalable way to read long records without reading every page by hand.
Catalog record
| CREATOR | Satya Narayana Cheetirala 1 , Ganesh Raut 1 , Dhavalkumar Patel 1 , Fabio Sanatana 1 , Robert Freeman 1 , Matthew Levin 2,4 , Girish N. Nadkarni 3,4 , Omar Dawkins 6 , Reba Miller 5 , Randolph M. Steinhagen 7 , Eyal Klang 3,4 , Prem Timsina 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-20320