Drawing Spec Compliance

AI-powered PDF drawing analysis pipeline for technical specification validation.

The system processes construction and engineering PDFs, separates individual drawings from multi-page albums, extracts text and visual information, retrieves relevant specification chunks with RAG, and generates structured compliance reports using local LLMs and deterministic validation rules.

Designed for offline/local workflows where auditability and controlled validation are more important than generic AI responses.

---

Features

• PDF drawing extraction and album splitting
• OCR with Tesseract + EasyOCR fallback
• Multimodal drawing understanding with LLaVA
• FAISS RAG retrieval over technical specifications
• Structured JSON compliance reports
• Deterministic rule-based validation
• Local Ollama-powered inference
• Multi-drawing PDF support
• Heuristic fallback when LLM is unavailable

---

Workflow

PDF / Album
      ↓
Drawing Detection
      ↓
Text + OCR Extraction
      ↓
Vision Analysis (LLaVA)
      ↓
RAG Retrieval (FAISS)
      ↓
LLM Compliance Check
      ↓
Rule Validation
      ↓
Structured JSON Report

---

Tech Stack

• Python
• PyMuPDF
• Tesseract OCR
• EasyOCR
• Ollama
• LLaVA
• Mistral
• FAISS
• sentence-transformers
• Pydantic

---

Example Use Cases

• Construction drawing validation
• Technical specification compliance
• Engineering document QC
• Supplier drawing review
• Automated document auditing

---

Installation

git clone https://github.com/YOUR_USERNAME/drawing-spec-compliance.git

cd drawing-spec-compliance

pip install -r requirements.txt

---

Build Specification Index

python -m scripts.index_rag --build-tz

---

Usage

Analyze single PDF:

python main.py --pdf drawing.pdf --drawing-id A-100

Analyze multi-page album:

python main.py --pipeline-album --pdf album.pdf

Check Ollama connection:

python main.py --check-ollama

---

Example Output

{
  "drawing_id": "A-101",
  "is_compliant": false,
  "issues": [
    {
      "field": "wall_thickness",
      "expected": "200mm",
      "actual": "160mm"
    }
  ]
}

---

Architecture

---

Project Structure

scripts/        # OCR, RAG, validation, LLM pipeline
config/         # Rules and validation settings
docs/           # Technical specifications
tests/          # Unit tests
data/           # Generated indexes and outputs

---

Notes

• Fully local/offline-capable pipeline
• No cloud APIs required
• Designed for auditable engineering workflows
• Deterministic validation layer independent from the LLM

---

License

MIT