geniAi-json-toolkit 🚀

A robust, zero-dependency JSON parser built specifically for LLM streaming, Python-isms, VLM spatial coordinates, and tool calls.

Handling JSON from Large Language Models (LLMs) and Vision-Language Models (VLMs) is notoriously painful. Models often hallucinate Python syntax, inject conversational Markdown, or stream incomplete objects that crash standard parsing libraries.

geniAi-json-toolkit is a drop-in toolkit that seamlessly repairs and normalizes unpredictable AI data into strict, usable Python objects in real-time.

✨ Features

• ⚡ Streaming & Partial Parsing: Auto-closes quotes, brackets, and dangling colons to parse incomplete JSON strings mid-stream.
• 🗣️ Yapping & Markdown Filter: Automatically strips conversational filler and markdown code blocks (e.g., ```json).
• 🐍 Python-ism Auto-Repair: Converts hallucinatory single quotes ('), True, False, and None into valid JSON equivalents.
• 👁️ Vision Coordinate Normalization: Auto-detects bounding boxes (0-1000 scales) from VLM outputs and rescales them to your physical camera resolution.
• 🛠️ Unified Tool Calling: Standardizes varying function-call formats (OpenAI, Anthropic) into a single predictable structure.
• 🛡️ Pydantic Safe-Cast: Streams partial data directly into Pydantic models without throwing AttributeError for missing fields.

📦 Installation

pip install geniAi-json-toolkit