错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Speeding up Government Procurement Workflows with LLMs

  • Aashna Sofat,
  • Balwinder Sodhi

摘要

Public procurement requires careful preparation of tender documents that must comply with regulatory requirements and the tendered items’ specifications. Later, when the bids are received, they need to be diligently evaluated for compliance. All these steps involve significant manual effort from the subject matter experts (SME). The bid evaluation alone takes several days to complete, often leading to delays and cost overruns. We present a system design that leverages LLMs to significantly reduce the bid analysis and evaluation effort. Traditional methods often struggle with extracting nuanced information and understanding complex relationships within bid documents. Our proposed system addresses these challenges by integrating LLMs for contextual understanding and retrieval-augmented generation for precise query answering. We present a comprehensive system design, highlighting its innovative features and potential for advancing information analysis tasks in tender bid evaluation. We have evaluated our system on diverse tenders data available through Indian e-procurement portals [2]. Our experiments show that our system is able to reduce the bid compliance check time from hours to seconds.