🤖 AI-Based DPR Evaluation System (SIH – MDoNER)

📌 Overview

The Ministry of Development of North Eastern Region (MDoNER) evaluates hundreds of Detailed Project Reports (DPRs) for infrastructure and socio-economic development projects. Manual evaluation is time-consuming, inconsistent, and prone to human error, often delaying project approvals.

This project implements an offline AI-powered DPR evaluation system that automatically extracts, analyzes, and evaluates DPR PDFs using NLP, OCR, and Retrieval-Augmented Generation (RAG) to support faster, consistent, and explainable decision-making.

---

🎯 Key Objectives

• Automate DPR evaluation for completeness, consistency, and risk
• Handle scanned, multilingual, and unstructured PDFs
• Provide explainable compliance validation
• Enable offline deployment for low-connectivity regions

---

🔄 System Workflow

1. DPR PDF upload via FastAPI
2. Hybrid PDF extraction (PyMuPDF → pdfplumber → OCR fallback)
3. Sentence-aware semantic chunking (200–500 words)
4. Retrieval-Augmented Generation (RAG) for context-aware analysis
5. Rule-based compliance and consistency checks
6. Evidence extraction from source text
7. Structured JSON output for decision support

---

✨ Core Features

• Hybrid PDF extraction for digital and scanned DPRs
• OCR fallback using PaddleOCR and Tesseract
• Semantic chunking preserving contextual information
• RAG-based context-aware analysis
• Deterministic rule-based compliance engine
• Offline-first design using local LLMs (Ollama)
• FastAPI backend with a single /process endpoint

---

🛠 Tech Stack

Main

• Python
• FastAPI
• NLP (RAG)
• Rule-Based Systems

Libraries & Tools

• PyMuPDF, pdfplumber
• PaddleOCR, pytesseract, Pillow
• Sentence Transformers
• Ollama
• NLTK

---

📊 Output

The system generates:

• Extracted text in JSON format
• Semantic text chunks
• Answers with evidence snippets
• Structured compliance and risk insights

---

👥 Team & Contribution

• Developed as part of Smart India Hackathon (SIH).