No comments yet.
I'm Erick, founder of cStructure. After 18 months of building in stealth, we're opening our causal inference platform for public beta. Last week we launched in the US, this week we are expanding to Canada.
Our story started when I was leading data science teams and kept hitting the same wall: causal analyses would fall apart in stakeholder meetings. Subject matter experts would ask why their favorite variable wasn't in the analysis. Data scientists would latch on to their favorite correlation-based approach (e.g. XGBoost + SHAP, even though SHAP's docs describe why it's the wrong tool for causal inference*). Leaders would question assumptions. What should have been knowledge-driven decisions turned into gridlock.
Everything changed when we started using causal diagrams. These simple visual maps of cause-and-effect became a shared language between experts, analysts, and decision-makers. Domain knowledge could be captured precisely. Assumptions became explicit. Teams could focus on the right questions and controls.
But building and validating these models was painful - scattered across whiteboards, papers, and custom code. We built cStructure to make rigorous causal inference collaborative and accessible.
Try it: https://cstructure.dev (free during public beta + simple demo canvas + guided tutorials by the first Sign Up button) Features: https://cstructure.dev/#features
Our team comes from life sciences, energy, and tech, where we've seen these methods scale from startups to major healthcare systems. The platform lets teams: - Kickstart and revise a causal diagram with AI assistance - Collaboratively build and validate causal models - Detect potential biases and validate assumptions - Connect models directly to data with automated checks - Run analyses in-browser using JupyterLite (no server needed) or export to your private environment
Technical details: - Built with React + Yjs + WASM
Our roadmap includes federated privacy-preserving learning, FAIR causal models, extensible analysis modules, and integration with knowledge graphs and scientific evidence.
Yes, there are still rough edges. But we're sharing now because we want to build this with feedback from the community. We want to understand what teams actually need to move beyond correlation-machines to doing real science.
Would love feedback from researchers, data scientists, and anyone interested in bridging the gap between domain expertise and statistical rigor. What would help your team adopt causal inference?
- Erick (at cstructure.io)
* https://shap.readthedocs.io/en/latest/example_notebooks/over...
