AIXC Research / Domains
Research Domains

Four Domains.
One Framework.

The same ACHT and VAP architectures adapt to radically different scientific domains through domain-specific encoders. Insights discovered in one field transfer to accelerate progress in the others.

DOMAIN 01

Drug Discovery & BioSciences

The Problem

Traditional drug discovery takes 10–15 years and $2.6B per approved molecule. Most candidates fail in late-stage clinical trials because early screening optimizes for binding affinity alone, ignoring causal mechanisms of action, off-target effects, and synthesizability constraints.

ACHT + VAP Solution

ACHT's Bayesian optimization explores chemical space 10–100x faster than grid search, while generative models propose novel scaffolds. Causal discovery on STRING v12.0 protein networks identifies mechanistic pathways. VAP's self-falsification catches false-positive binding predictions before they waste wet-lab resources.

BindCraft STRING v12.0 RDKit DeepChem AlphaFold

See full platform at BioSciences site

3,529
Novel candidates generated from 100K+ protein interactions
STRING v12.0 → ACHT Pipeline
10–100x
Faster hypothesis exploration vs. conventional screening
Bayesian Optimization speedup
DOMAIN 02

Financial AI & Causal Economics

The Problem

Financial models confuse correlation with causation, leading to brittle strategies that collapse under regime change. Autonomous trading systems amplify errors through echo chambers where multiple AI agents converge on identical wrong conclusions.

ACHT + VAP Solution

ACHT discovers causal factor structures in financial time series, distinguishing genuine economic drivers from spurious correlations. VAP's polyculture agents maintain genuine diversity in trading strategies, while echo chamber detection prevents correlated failure modes. Self-falsification stress-tests strategies against historical regime changes.

Causal Factors Regime Detection Echo Chamber Prevention Multi-Agent Trading
23%
False positives prevented by echo chamber detection
VAP Polyculture Agents
4.2x
Improvement in out-of-distribution robustness
Causal vs. correlational models
DOMAIN 03

Legal Reasoning & Neuro-Symbolic AI

The Problem

Legal reasoning requires both statistical pattern recognition and formal logical structure. Pure neural approaches hallucinate precedents. Pure symbolic approaches cannot scale to the volume and ambiguity of real-world legal corpora.

ACHT + VAP Solution

ACHT's causal discovery identifies precedent chains and statutory dependencies. Neuro-symbolic integration embeds formal logic constraints into neural architectures, ensuring generated arguments are both statistically grounded and logically valid. VAP verifies argument consistency against the full statutory graph.

Neuro-Symbolic Precedent Graphs Formal Logic Argumentation Mining
0
Hallucinated precedents in verified outputs
VAP Self-Falsification
94.7%
Logical consistency in generated arguments
Neuro-Symbolic hybrid
DOMAIN 04

Materials Science & Ceramics

The Problem

Advanced ceramics design requires navigating a combinatorial composition space where small changes in stoichiometry produce dramatic property changes. Experimental exploration is slow and expensive, with each synthesis-characterization cycle taking weeks.

ACHT + VAP Solution

ACHT's Bayesian optimization navigates the ceramic composition space, with physics-informed surrogates encoding thermodynamic constraints and phase stability rules. Generative models propose novel compositions satisfying target property profiles. VAP validates predictions against conservation laws and known phase diagrams before committing to synthesis.

Phase Diagrams Composition Optimization PINNs PyTorch Geometric Sakana Platform
3,000x
Faster than manual experimental exploration
AI-Scientist Engine acceleration
87%
Synthesis success rate for AI-proposed compositions
VAP physics validation
Cross-Domain Transfer

Synergy Across Domains

The deepest insight of ACHT is that causal structures discovered in one domain often have structural analogs in others. Knowledge transfer across domains is not metaphorical — it is mathematical.

Bio→Finance
Network Topology Transfer
Protein interaction network analysis methods directly apply to financial contagion modeling. The same community detection algorithms that identify functional protein modules identify correlated asset clusters.
Finance→Legal
Causal Chain Reasoning
Causal factor models from finance provide the mathematical framework for analyzing chains of legal causation. Counterfactual reasoning techniques transfer directly to legal liability analysis.
Materials→Bio
Physics-Informed Constraints
Physics-informed neural networks developed for materials thermodynamics adapt to enforce biochemical constraints in drug design: conservation of charge, steric compatibility, and thermodynamic binding stability.
Legal→Materials
Formal Verification Methods
Formal logic verification developed for legal argument consistency provides the mathematical backbone for verifying that AI-proposed material compositions satisfy hard physical constraints and safety requirements.
Bio→Materials
Generative Molecular Design
Diffusion models trained on molecular structures adapt to generate novel ceramic compositions by treating crystal structures as 3D molecular graphs with periodic boundary conditions.
Finance→Bio
Portfolio Optimization
Multi-objective portfolio optimization from finance maps directly to drug candidate portfolio selection: balancing risk (toxicity), return (efficacy), and diversification (mechanism coverage).
Next Step

Explore Publications

23+ peer-reviewed papers across all four domains. Filter by topic, venue, or architecture.

View All Publications Back to Architectures