🧠 Sankalp AI

AI/ML-Based Auto Evaluation of R&D Proposals for Ministry of Coal

Sankalp AI is an intelligent proposal evaluation platform designed to assist the Ministry of Coal (S&T Department) in automatically assessing, scoring, and ranking R&D project proposals using advanced AI, NLP, and RAG pipelines.

It streamlines the end-to-end evaluation process by analyzing novelty, financial feasibility, technical strength, and relevance, while providing a powerful dashboard for evaluators.

---

🚀 Tech Stack

Frontend: React, Tailwind CSS
Backend: Node.js, Express, FastAPI, Python
Databases: MongoDB, PostgreSQL + PGVector
AI/ML: Language Transformers (all-mpnet-base-v2), RAG Pipeline , LLMs Architecture: Microservices
Other: PyPDF, OCR (planned), Speech-to-Text, Text-to-Speech

---

🎯 Problem Statement

Manual evaluation of R&D proposals is:

• Time-consuming
• Subjective
• Hard to benchmark at scale

Sankalp AI automates this process to ensure: ✅ Fairness
✅ Consistency
✅ Speed
✅ Data-driven decisions

---

🖼️ App Demo

🏠 Dashboard Overview

📂 All Proposals View

📑 Proposal Scorecard

🎙️ AI Voice Agent

🎯 Relevance Evaluation

🛠️ Technical Evaluation

✨ Novelty Evaluation

💰 Financial Evaluation

---

✨ Key Features

📄 1. Document Processing

• Extracts text from newly submitted proposals (PDF → Text).
• Uses PyPDF for text extraction.
• Planned support for OCR for scanned documents.

---

🔍 2. Novelty Checker (Originality Score)

• Splits proposal text into chunks (~8000 characters).
• Generates embeddings using all-mpnet-base-v2 transformer model.
• Stores embeddings in PostgreSQL with PGVector.
• Performs semantic similarity search against past proposals.
• Returns a Novelty Score (0–100) based on uniqueness.

Example: “Father’s father’s wife” ≈ “Grandmother” — semantic understanding beyond keywords.

---

💰 3. Financial Validator

Based on S&T Funding Guidelines (Ministry of Coal).

Evaluates and scores:

1. Budget justification & correctness
2. Cost-benefit balance
3. Commercialization potential
4. Sustainability post-completion
5. Financial risk identification

Uses custom-trained LLMs to flag violations and risks automatically.

---

⚙️ 4. Technical & Relevance Scoring

Technical Score Criteria:

• Practicality of approach
• Proof-of-concept strength
• Resource availability
• Risk mitigation plans
• Timeline feasibility

Relevance Score Criteria:

• Applicability to coal industry problems
• Alignment with S&T-PRISM & ministry priorities
• Safety & environmental impact
• Efficiency & cost reduction
• Adoptability by PSUs (CIL, SCCL, CMPDI, etc.)

---

🏆 5. Overall Scoring & Ranking

• Combines Novelty, Financial, Technical, Relevance scores.
• Computes Overall Score (out of 10).
• Automatically ranks proposals.
• Updates live benchmarks for evaluators.

---

📊 6. Evaluator Dashboard

• 📌 Welcome overview & latest submissions
• 📂 All applications with search, filter & sort
• 📑 Individual proposal scorecards with detailed analysis
• 🎙️ AI Voice Agent acting as a co-evaluator

---

🏗️ System Architecture

We follow a Microservices Architecture for scalability and modularity.

🧩 Microservices:

1. Novelty Service

   • Python backend
   • Transformer model
   • PostgreSQL + PGVector

2. Financial & Technical Scoring Service

   • FastAPI, Python
   • Gemini LLMs

3. Web Application

   • MERN Stack (MongoDB, Express, React, Node)
   • Tailwind CSS UI

4. Voice Agent Service

   • Speech-to-Text
   • Text-to-Speech
   • LLM-powered responses