UM CAEN

Project Overview

When submitting grant proposals to the National Science Foundation (NSF), researchers must comply with the PAPPG, which outlines strict guidelines for the proposal formatting and proposal content. In addition to these requirements, there are often research-specific guidelines that the proposal must adhere to.

Currently, the compliance review process is done manually by the College of Engineering and requires expert review, manual verification, and significant time and effort.

Approach

Our project aims to build an AI-powered compliance assistant that automatically checks uploaded proposals against both the PAPPG and any proposal-specific requirements. This tool is designed to streamline the review process for both researchers, ADRs (Associate Deans of Research), and the Office of Research and Sponsored Projects (ORSP), ultimately resulting in a lower rejection rate.

Roles & Responsibilities

Building an end-to-end AI-powered grant compliance tool that automates proposal evaluation against NSF guidelines using GPT-4, LangChain, and a Retrieval-Augmented Generation (RAG) pipeline

Developing a scalable, cloud-based web application using Flask and Vue 3 with TypeScript, with backend services containerized via Docker and deployed using AWS Fargate and Lambda for serverless infrastructure

Leading Agile sprint planning using Jira, coordinated cross-functional development timelines, and implemented CI/CD pipelines to streamline deployment and improve team velocity by 30%

Implementing context-aware querying and retrieval workflows combined with prompt engineering techniques to extract relevant compliance rules from documents, increasing response precision and reducing manual review time

Technologies/Tools Used:

Python, TypeScript, Flask, Docker, AWS Lambda, S3 & Fargate, Vue, LangChain, Open AI GPT-4o, RAG Pipeline, Prompt Engineering, Google O-Auth