REQUIREMENTS
This section outlines the comprehensive requirements for our systematic review generation system, including project goals, user needs, functional specifications, and prioritized feature sets.
Project Overview
Our client CarefulAI focuses on creating safe and reliable AI agents for various fields, including healthcare and economic development. They manage AI safety and assurance across the complete AI supply chain, from algorithm design to global management.
Our project, in partnership with Joseph Connor from CarefulAI, is developing a RAG system to undertake systematic reviews, serving as an important tool for NHS integration. This initiative aligns with CarefulAI's values and guides us toward creating AI applications that benefit humanity.
Project Goals
Medical Paper Analysis
Develop RAG system for comprehensive medical paper analysis
Automated Review
Generate structured systematic reviews automatically
Workload Reduction
Achieve 50%+ efficiency improvement for medical professionals
PDF Processing
Enable PDF upload & natural language query processing
Requirement Gathering
Client Requirements
- Web application with PDF processing pipeline
- Vector database integration (Pinecone)
- Matplotlib visualization for quality assessments
- Systematic review generation following PRISMA guidelines
- User authentication and history tracking
User Survey Highlights
User Personas
Dr. Sarah Johnson
Medical Researcher
- Conducts systematic reviews monthly
- Needs to stay current with latest research
- Frustrated with time-consuming manual reviews
Prof. Michael Chen
Academic Supervisor
- Oversees multiple research projects
- Needs to validate student literature reviews
- Wants consistent methodology across reviews
Use Cases
| ID | Use Case | Flow | Result |
|---|---|---|---|
| UC01 | Log in | 1. User enters credentials 2. System verifies login |
User is logged in or sees error |
| UC02 | Sign up | 1. User enters details 2. System registers user |
Account created or error shown |
| UC03 | Upload PDFs | 1. User selects files 2. System processes upload |
Files ready for analysis |
| UC04 | Generate Review | 1. Retrieve data 2. Send to LLM 3. Generate sections |
Structured review created |
| UC05 | Download Results | 1. User requests download 2. System generates PDF |
User receives report |
Requirements Prioritization
Functional Requirements
Must Have
- User can input PDFs
- User can enter a prompt to define the systematic review
- Systematic review is displayed to the user
- Vector database integration
- LLM-powered section generation
Should Have
- User authentication system
- Review history tracking for users
- Quality assessment visualisation of the review through graphs
Could Have
- Users can view the source files of each review
- PDFs of the systematic review can be exported
- Medical paper database integration
Won't Have
- Collaborative features with multiple users
- Allow users to input a prompt through speech
Non-Functional Requirements
Must Have
- Web application implementation
- RAG architecture compliance
- Data encryption (bcrypt) for user authentication
Should Have
- High performance standards
- Maintainable codebase
Won't Have
- Cross-platform compatibility
- Multilingual support