🧠 CausalIQ Knowledge Project

The CausalIQ Knowledge project represents a novel approach to causal discovery by combining the traditional statistical structure learning algorithms with the contextual understanding and reasoning capabilities of Large Language Models. This integration enables more interpretable, domain-aware, and human-friendly causal discovery workflows.

Current Version: v0.6.0

Quick Links:

• 📖 Full Documentation
• 💻 Repository

Key Features

🧠 LLM Graph Generation (generate_graph)

Generate causal PDGs from network context specifications using any of 7 LLM providers (OpenAI, Anthropic, Groq, DeepSeek, Mistral, Gemini, Ollama), with multi-sampling via llm_seed and semantic variable name disguising. Available as CLI and workflow action.

💾 Response Caching

SQLite-based query and response caching for stable, reproducible LLM experiments, with provider-specific limits, CLI cache management tools, and automatic recovery from malformed LLM responses.

🔌 Direct Vendor API Clients

Lightweight LLM integration using direct vendor APIs via httpx — no LiteLLM or LangChain wrappers — for full control and easy debugging.

Future Directions

🤝 Human Engagement

• Natural language constraints: Specify domain knowledge in plain English
• Expert knowledge incorporation by converting expert understanding into algorithmic constraints
• LLMs convert natural language questions to causal queries
• Interactive causal discovery where structure learning or LLMs identify areas of causal uncertainty and can test causal hypotheses through dialogue

🪟 Transparency and Interpretability

• LLMs interpret structure learning process and outputs, including their uncertainties
• LLMs interpret causal inference results including uncertainties
• Contextual graph interpretation to explain variable meanings and relationships
• Uncertainty communication with clear explanation of confidence levels and limitations
• Report generation including automated research summaries and methodology descriptions

🔒 Stability and Reproducibility

• Stable randomisation of e.g. data sub-sampling
• Evaluation and development of simple context-adapted LLMs

Integration with Ecosystem

• 🔍 CausalIQ Discovery makes use of this package to learn more accurate graphs.
• 📊 CausalIQ Analysis uses graph averaging from causaliq-analysis to resolve uncertain edges identified by this package.
• 🔮 CausalIQ WhatIf uses this package to explain predictions made by learnt models.
• 🤖 CausalIQ Workflow executes graph generation as workflow steps with cache integration.

---

CausalIQ Knowledge bridges the gap between statistical rigour and domain expertise, enabling more intelligent, interpretable, and human-collaborative approaches to causal discovery.