This repository documents the design and architecture of the NAAC Automation Portal, a proposed AI-powered compliance automation platform for the NAAC accreditation process. It outlines how artificial intelligence, large language models, document processing, and RAG-based chatbot systems can be integrated to streamline documentation, validation, and communication with the Internal Quality Assurance Cell (IQAC).
The NAAC Automation Portal provides an AI-based backend and web-based frontend to handle key stages in the accreditation workflow:
- Excel metadata validation and proof document checking
- Centralized aggregation of documents from various campuses
- RAG-based chatbot for user assistance on NAAC criteria
- OCR support for scanned documents
- Summarization, document comparison, and missing data identification
The portal is designed to reduce manual effort, minimize submission errors, and improve the efficiency of preparing for NAAC accreditation.
The NAAC Automation Portal is structured as a modular, end-to-end automation pipeline designed to handle the entire documentation process for NAAC accreditation. It combines document ingestion, AI-based analysis, real-time interaction, and cloud storage integration.
-
Developed using Streamlit, with HTML/CSS/JavaScript components for interactivity
-
Provides functionality for:
- Campus and criterion selection
- File uploads (Excel and PDFs)
- Template viewing and editing (via Handsontable)
- Interaction with the AI chatbot
- Triggering summarization, document comparison, and missing information checks
Built on Python (Flask or FastAPI) and Node.js services, the backend handles:
- File processing (validation, merging, extraction)
- Session handling and routing requests from frontend
- Query dispatching to document databases and LLMs
- Summarization and NLP-based insights
- Utilizes Tesseract OCR to extract machine-readable text from scanned PDFs or image-based documents
- Converts uploaded files into
.txtformat for consistency - Supports page-wise PDF conversion using
pdf2image, preprocessing, and noise filtering
-
Converts cleaned
.txtdocuments into semantic vector embeddings -
Pipeline steps:
- DirectoryLoader: Recursively loads documents from campus/criterion directories
- RecursiveCharacterTextSplitter: Divides large documents into overlapping chunks
- OpenAIEmbeddings: Generates semantic embeddings from text
- FAISS: Indexes embeddings into a local searchable vector database
-
Central AI agent that processes user queries contextually
-
Combines:
- LangChain’s
AgentExecutorto orchestrate tools - Tools like
WikipediaQueryRun,GoogleSearchAPIWrapper - A retriever that queries the FAISS vector store using semantic similarity
- LangChain’s
-
The agent selects between internal data, Wikipedia, and web search based on the query
-
Real-time assistant built on OpenAI GPT models
-
Responds to:
- Queries about NAAC criteria
- Missing template data
- Process flow clarification
-
Connects to the document embeddings, so answers are context-aware and grounded in uploaded documentation
-
Implements a Map-Reduce strategy using LangChain:
- Map Phase: Generates summaries for each chunk
- Reduce Phase: Merges intermediate summaries into a cohesive final summary
-
Ensures scalable summarization for large datasets (e.g., 80+ documents per criterion)
-
Accepts multiple document uploads
-
Runs a vector-based semantic alignment and comparison
-
Detects redundant sections, differences, and structural mismatches
-
Outputs:
- Aligned text comparisons
- Highlighted missing data
- Merged documents with contextual alignment
-
Accepts a base (parent) document and a partially filled template
-
Identifies missing fields via:
- Embedding comparison
- Query chains
- Prompt engineering (e.g., “Compare X with Y, fill in Z”)
-
Returns either text or completed templates
-
Uses Node.js/Express for:
- Directory creation per campus and criterion
- File downloads/uploads
-
Supports OneDrive API integration for cloud aggregation of proof documents
The system provides a comprehensive feature set that transforms the traditionally manual and error-prone NAAC compliance process into an intelligent, efficient, and automated workflow. Each of the results mentioned are explained in detail in the document above with picture proofs representing the functionality.
-
Excel templates are parsed and checked for:
- Metadata completeness
- Matching document names and formats
- Required fields for each NAAC metric
-
Immediate feedback is given through the frontend
-
Documents are sorted automatically by:
- Campus
- Criterion
- Document type (Proof, Template, etc.)
-
Integrated into OneDrive for persistent storage
- Context-aware Q&A over uploaded documents
- Answers template-specific questions (e.g., “What is missing in 1.1.1?”)
- Uses LLM with vector search and fallback to web search tools
-
Handles multi-document summarization
-
Outputs include:
- Summary by criterion
- Aggregated summaries for full submissions
-
Designed for IQAC reporting and review
-
Converts scanned PDFs into text
-
Supports:
- English documents with complex layouts
- Table and column recognition (basic)
- Compares input files with parent documents
- Finds unfilled fields or mismatched entries
- Suggests or fills data automatically using LLMs
-
Compares multiple documents chunk-by-chunk
-
Identifies:
- Overlapping content
- Deviations in structure
- Additional or missing sections
-
Produces consolidated output
-
Handsontable-based Excel preview and editing
-
Users can:
- Edit directly in browser
- Download updated versions
- Save progress locally during sessions
-
LangChain tool-based architecture
-
Easily extendable with:
- New NAAC criteria
- Additional agents (e.g., CV parser, LMS sync tool)
- Custom summarizers or validators
- Python 3.11
- Node.js, Express
- Flask or FastAPI
- OpenAI API
- FAISS for vector storage
- LangChain
- Tesseract OCR
- Streamlit
- HTML5, CSS3, JavaScript
- Handsontable.js
- Bootstrap 4+
This repository contains documentation only. The steps below describe how to set up the system if implementing the described architecture.
Install dependencies:
pip install langchain openai streamlit faiss-cpu pytesseract pdf2image PyPDF2
pip install python-dotenv sse_starlette bs4 wikipediaInstall Tesseract OCR and poppler for PDF conversion:
-
Ubuntu/Debian:
sudo apt install tesseract-ocr poppler-utils
-
Windows:
- Download Tesseract OCR
- Add the executable path to the system environment variables
Install dependencies:
npm install express node-fetch fs corsUse Node.js scripts to manage file uploads/downloads and directory structure.
Configure the following keys in a .env file:
OPENAI_API_KEY=your_openai_key
GOOGLE_API_KEY=your_google_api_key
GOOGLE_CSE_ID=your_custom_search_id
LANGCHAIN_API_KEY=your_langchain_api_key
Start the frontend:
streamlit run app.pyStart the backend (if used):
node index.js- Navigate to the Streamlit frontend
- Select NAAC criterion and upload Excel/PDF files
- Validate templates or use chatbot for questions
- Run document comparison or summarization tools
- Export outputs (text or PDF reports)
This repository includes:
NAAC Automation.docx: Complete project documentation- System architecture, block diagrams, and design flow
- Details on modules such as RAG chatbot, OCR, vector database, and summarization