TeleOCR is a lightweight 1.2B parameter Vision-Language Model that replaces commercial cloud Document AI services. It parses camera-captured receipts, contracts, and math tables directly into structured JSON and SQLite databases offline.
| Step 1: Input | Step 2: AI Action | Step 3: Result |
|---|---|---|
| 📷 Photographed Invoice / Contract | 🧠 TeleOCR Geometry & OTSL Grid Parsing | 📊 Structured JSON & SQLite Database |
- 🏆 OmniDocBench #1 Leader: Achieves 96.87 overall score, outperforming GPT-5.2 (86.52) and Gemini 3 Pro (92.85).
- 📷 Dewarping-Free Camera OCR: Parses angled, crumpled, or unevenly lit phone photos of documents without external dewarping tools.
- 📊 Table Grid Decoupling: Converts complex nested data tables directly into HTML grid structures using One-Table-Structure-Language (OTSL).
- 🔒 100% Offline Autonomy: All document extraction runs locally on your machine with zero recurring cloud API fees or privacy risks.
TeleOCR/
├── main.py
├── requirements.txt
├── README.md
└── tests/
└── test_main.py
Copy and paste the following single PowerShell command to install the TeleOCR dependencies:
pip install transformers torch pillowTo download the TeleOCR model weights directly from Hugging Face:
hf download StarDoc-AI/TeleOCR --local-dir models/TeleOCRRun the automated document parsing loop across all input files:
python main.pyThe script automatically scans all document images in inputs/, runs the TeleOCR parsing loop, saves records to SQLite (outputs/documents.db), and builds the final Markdown report (outputs/outputs.md).
- Receipt & Expense Automation: Convert phone photos of paper receipts into structured JSON for accounting software.
- Contract Clause Extraction: Extract multi-column legal terms and nested billing schedules into local databases.
- STEM Academic Digitization: Parse research papers to convert mathematical equations into LaTeX code blocks.
- Supply Chain Invoice Processing: Parse messy shipping manifests and warehouse tables into structured database entries.
- Private On-Premise Archiving: Process sensitive financial and medical documents without uploading data to public cloud APIs.
- GUI Dashboard Interface: Interactive desktop dashboard for drag-and-drop document upload and batch JSON preview.
- Multi-Language OCR Expansion: Native support for non-Latin scripts and low-resource historical manuscripts.
- Automated Vector Embedding Pipeline: Direct integration with local vector databases for RAG document search.
- Mobile Camera App Integration: On-device camera scanning powered by quantized GGUF models.
- Custom Schema Mapper: User-defined JSON schema mapping for enterprise ERP software integrations.
TeleOCR Vision-Language Model Document AI OTSL Offline OCR Receipt Parsing Hugging Face Models Local AI Model Python Document Processing SQLite Export