Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Research Paper AI Assistant

An intelligent research management system built on Microsoft Power Platform that automatically analyzes, classifies, and cross-references 100+ academic papers. Features RAG-powered Q&A with multi-agent evaluation, contradiction detection across papers, proactive weekly insights via Microsoft Graph, and full governance controls.

Architecture

                        ┌──────────────────────────────┐
                        │   SharePoint Document Library │
                        │   (100+ Research Paper PDFs)  │
                        └──────────────┬───────────────┘
                                       │ Upload trigger
                                       ▼
                    ┌──────────────────────────────────────┐
                    │   Power Automate: Ingestion Pipeline  │
                    │                                      │
                    │  1. Azure Document Intelligence       │
                    │     └─► Extract text from PDF         │
                    │  2. Text Chunking (500 tokens, 50     │
                    │     token overlap)                    │
                    │  3. OpenAI Embeddings API             │
                    │     └─► Vectorize each chunk          │
                    │  4. Auto-Classification Agent         │
                    │     └─► Topic, tags, methodology      │
                    │  5. Contradiction Detection Agent     │
                    │     └─► Cross-reference with existing │
                    │  6. Azure AI Search                   │
                    │     └─► Index vectors + metadata      │
                    │  7. Dataverse                         │
                    │     └─► Log ingestion status          │
                    └──────────────────────────────────────┘

                    ┌──────────────────────────────────────┐
                    │   Power Automate: RAG Query Pipeline  │
                    │                                      │
                    │  1. Receive query from Copilot Studio │
                    │  2. Embed query via OpenAI API        │
                    │  3. Vector search in Azure AI Search  │
                    │  4. Load prompt template from         │
                    │     Dataverse (versioned)             │
                    │  5. Generate answer with citations    │
                    │  6. Evaluation Agent                  │
                    │     └─► Faithfulness check            │
                    │     └─► Citation accuracy check       │
                    │     └─► Confidence scoring            │
                    │  7. Log query + tokens to Dataverse   │
                    └──────────────────────────────────────┘

                    ┌──────────────────────────────────────┐
                    │   Copilot Studio (Teams)              │
                    │                                      │
                    │  Topics:                             │
                    │  - Ask about research (RAG Q&A)      │
                    │  - Summarize a paper                 │
                    │  - Compare papers                    │
                    │  - Show contradictions               │
                    └──────────────────────────────────────┘

                    ┌──────────────────────────────────────┐
                    │   Power Automate: Weekly Digest       │
                    │                                      │
                    │  Scheduled (every Monday 9 AM)       │
                    │  1. Query Dataverse for weekly stats  │
                    │  2. Identify research gaps            │
                    │  3. Find cross-paper connections      │
                    │  4. Send digest via Microsoft Graph   │
                    │     (Outlook email)                   │
                    └──────────────────────────────────────┘

                    ┌──────────────────────────────────────┐
                    │   Power Apps: Admin Dashboard         │
                    │                                      │
                    │  - Paper management + metadata        │
                    │  - Prompt version management (A/B)    │
                    │  - Cost/token usage tracking          │
                    │  - Query analytics + feedback review  │
                    │  - Contradiction alerts               │
                    └──────────────────────────────────────┘

                    ┌──────────────────────────────────────┐
                    │   Security & Governance               │
                    │                                      │
                    │  - Entra ID SSO                      │
                    │  - RBAC (Admin / Reader roles)        │
                    │  - Azure Key Vault (API secrets)      │
                    │  - DLP Connector Policies             │
                    │  - Dev/Prod environment separation    │
                    │  - Solution packaging (ALM)           │
                    └──────────────────────────────────────┘

Tech Stack

Layer Technology
Document Storage SharePoint Online
Workflow Automation Power Automate (Cloud Flows)
Chat Interface Copilot Studio (deployed in Teams)
Admin UI Power Apps (Canvas App)
Data Layer Dataverse
PDF Extraction Azure Document Intelligence (Form Recognizer)
Vector Search Azure AI Search (Free tier)
LLM OpenAI API (gpt-4o-mini + text-embedding-3-small)
Email Integration Microsoft Graph API (Outlook)
Security Entra ID, Azure Key Vault, DLP Policies

Key Features

Intelligent Ingestion (beyond basic chunking)

  • Auto-classification of paper topic, methodology, and domain
  • Auto-extraction of key findings, techniques, and datasets
  • Contradiction detection against existing papers in the library
  • Knowledge tagging for cross-referencing

RAG with Evaluation (beyond basic Q&A)

  • Multi-agent pipeline: generate answer → evaluate faithfulness → score confidence
  • Citation accuracy verification before returning to user
  • Versioned prompt templates stored in Dataverse with A/B testing support

Proactive Insights (beyond reactive chat)

  • Weekly digest email via Microsoft Graph with:
    • Research gap analysis
    • Cross-paper connection suggestions
    • Trending query topics
    • Collection health metrics

Governance & Observability

  • Prompt versioning with performance tracking
  • Per-query cost/token logging
  • Response quality metrics over time
  • Entra ID auth, RBAC, Key Vault, DLP policies
  • Dev/Prod ALM with solution packaging

Project Structure

ResearchPaperAssistant/
├── README.md
├── docs/                    # Setup guides and documentation
├── prompts/                 # All LLM prompt templates
├── flows/                   # Power Automate flow definitions (exported JSON)
├── power-apps/              # Power Apps source files
├── copilot-studio/          # Copilot Studio topic definitions
├── dataverse-schemas/       # Table schemas and relationships
├── scripts/                 # Utility scripts (bulk upload, testing)
└── tests/                   # Test cases and validation scripts

Azure Free Tier Resources

Service Free Allowance Usage
Form Recognizer 500 txns/month PDF text extraction
AI Search (F tier) 50 MB, 10K docs Vector index
Key Vault 10,000 ops API key storage
Cosmos DB 25 GB + 100 RU/s Backup store
Blob Storage 5 GB Staging area

Estimated total cost: $2-5 (OpenAI API usage from existing $10 credits)

Setup Guide

See docs/setup-guide.md for step-by-step instructions.

About

Research management system on Microsoft Power Platform: RAG Q&A, contradiction detection, and weekly insights across 100+ papers

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors