Causalit builds heterogeneous AI systems that combine large language models with causal reasoning to create explainable, edge‑deployable models for healthcare. By extracting cause‑and‑effect graphs from health data and training a compact Savant Language Model to reason over them, Causalit delivers fast, privacy‑preserving predictions with built‑in episodic memory for reliable decision support.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI models, especially large language models, rely on statistical correlations, which can lead to unreliable predictions and a lack of explainability. This limits their suitability for high‑risk domains such as healthcare, where understanding the reasoning behind a decision is essential.
Solution
Causalit builds heterogeneous AI systems that integrate causal reasoning with large language models. The platform first constructs a structured map of cause‑and‑effect relationships from diverse health data, then trains a compact Savant Language Model (sLM) to reason over this causal graph. Because the sLM operates on explicit causal knowledge, it can provide explanations for its conclusions, not just predictions. The resulting models are small enough to run on edge devices—clinical workstations or personal health hardware—eliminating the need for constant cloud connectivity and enhancing privacy, speed, and cost efficiency.
Target Audience
Primary customers are healthcare providers, medical device manufacturers, and health‑tech developers that require trustworthy, explainable AI for diagnostics, treatment recommendation, or patient monitoring.
Features
- Two‑stage pipeline that extracts causal relationships from heterogeneous health data and encodes them into a structured graph
- Savant Language Model (sLM) that reasons over the causal graph to generate explainable predictions
- Edge‑deployable architecture allowing inference on local devices without reliance on cloud services
- Built‑in episodic memory component that retains and utilizes past interactions for more reliable decision making
- Reduced model size compared to generic LLMs, enabling faster response times and lower computational costs