⚡ SAPIEN RAGBot

Sustainable • Dystopian • Mechanical Knowledge Retrieval

Status: Active Development

An industrial-grade RAG (Retrieval-Augmented Generation) system for welding defect diagnosis, powered by semantic search and confidence scoring. Built for mechanical engineers who need reliable, traceable answers about welding failures, root causes, and remediation strategies.

SAPIEN RAGBot combines FAISS-powered vector similarity search with multi-dimensional confidence metrics to deliver precise defect matching with structured metadata including symptoms, causes, prevention methods, and remedies.

View on GitHub Watch Demo Video

🌟 Key Features

🔍 Semantic Search

FAISS-powered vector similarity for precise defect matching

🧠 Confidence Scoring

Multi-dimensional confidence metrics

📊 Rich Metadata

Structured defect data with full traceability

🎨 Dystopian UI

Futuristic Streamlit interface with neon aesthetics

⚡ Fast API

Production-ready FastAPI backend

🔧 Mistral AI

🏗️ System Architecture

┌─────────────────────────────────────────────────────────────┐
│                     SAPIEN RAGBot System                    │
└─────────────────────────────────────────────────────────────┘
                              │
          ┌───────────────────┼───────────────────┐
          │                   │                   │
    ┌─────▼─────┐      ┌──────▼──────┐     ┌─────▼─────┐
    │ Frontend  │      │   Backend   │     │ Knowledge │
    │ Streamlit │◄────►│   FastAPI   │◄───►│   Base    │
    └───────────┘      └─────────────┘     └───────────┘
          │                   │                   │
    • Dystopian UI      • Vector Search     • FAISS Index
    • Query Input       • Confidence        • Embeddings
    • Results Display   • REST API          • Metadata

📊 Data Pipeline

PDFs → Text Extraction → Structured Chunking → Embeddings → FAISS Index
                              ↓
                        Mistral AI
                              ↓
                   Defect Name, Symptoms,
                   Causes, Prevention, etc.

Pipeline Stages:

Stage 1: Extract text from PDF welding manuals using pdfplumber
Stage 2: Extract structured defect data using Mistral AI
Stage 3: Generate AI-powered summaries for each defect
Stage 4: Create embeddings with Mistral embeddings model
Stage 5: Build FAISS vector index for fast similarity search

🔧 Technology Stack

Backend Framework FastAPI

Frontend Streamlit

Vector Search FAISS

AI Model Mistral AI

PDF Processing pdfplumber

Data Validation Pydantic

🧠 Confidence Scoring System

The system uses a multi-dimensional confidence model to ensure reliable defect diagnosis:

1. Similarity Confidence (40%)

Measures vector similarity using exponential decay of L2 distances:

sim_scores = np.exp(-distances)

2. Consensus Confidence (30%)

Measures agreement across retrieved documents:

consensus = 1.0 - (unique_defects - 1) / total_defects

3. Completeness Confidence (30%)

Checks field completeness (6 required fields):

completeness = filled_fields / 6

Final Score Calculation

final = (similarity × 0.4) + (consensus × 0.3) + (completeness × 0.3)

Confidence Levels

HIGH: ≥ 0.75 MEDIUM: 0.50 - 0.74 LOW: < 0.50

📡 API Example

Request

curl -X POST "http://127.0.0.1:8000/ask" \
  -H "Content-Type: application/json" \
  -d '{"query": "What causes porosity in aluminum TIG welding?"}'
      

Response

{
  "answer": {
    "defect_name": "Porosity",
    "welding_process": "TIG",
    "symptoms": "Small cavities in weld metal",
    "root_causes": "Contaminated shielding gas, dirty base metal",
    "prevention": "Use pure argon, clean metal surface",
    "remedies": "Grind out and re-weld affected areas",
    "material": "Aluminum",
    "position": "All",
    "source_doc": "AluminumWelding.txt",
    "summary": "Porosity in aluminum TIG welding..."
  },
  "confidence": {
    "confidence_score": 0.782,
    "similarity_confidence": 0.856,
    "consensus_confidence": 0.667,
    "completeness_confidence": 0.833,
    "confidence_level": "HIGH"
  },
  "sources": ["AluminumWelding.txt"]
}
      

🎨 User Interface

Experience the dystopian, futuristic interface of SAPIEN RAGBot with neon aesthetics and real-time animations.

SAPIEN RAGBot Interface - Query Terminal

SAPIEN RAGBot Interface - Results Display

💡 Use Cases

1. Quality Control

Quickly diagnose welding defects during inspection and identify root causes.

2. Training

Educate new welders about common defects, their causes, and prevention methods.

3. Maintenance Planning

Aggregate historical failure data to predict and prevent future issues.

4. Documentation

Centralize welding knowledge from multiple manuals into a searchable system.

🚧 Roadmap

✓ Multi-modal support (images, diagrams)
✓ Real-time re-ranking with cross-encoders
✓ User feedback loop for continuous improvement
⚡ Export reports to PDF
⚡ Multi-language support
⚡ Integration with PLM/ERP systems
⚡ Mobile app version

⚡ Built with precision for sustainable mechanical excellence ⚡

Low-key, dystopian, futuristic vibes activated 🔋