Technology Platform

TheraJect’s platform is an integrated AI-driven system for rare-earth innovation, combining materials discovery, bio-extraction, and digital twin supply-chain intelligence.

At its core is a unified AI inference and optimization framework that enables rapid exploration, evaluation, and deployment of candidate materials and processes across the rare-earth lifecycle.

Platform Architecture

TheraJect’s technology platform consists of three tightly integrated layers:

AI Inference Engine → Physics & Biology Modeling → System-Level Optimization

This architecture enables end-to-end innovation from atomic-scale materials design to system-scale supply-chain optimization.

1️⃣ AI/ML Inference Engine

Our AI layer enables inference-time discovery and decision-making, not just offline training.

It integrates:

  • Generative models (VAE, diffusion, GFlowNet) for candidate generation
  • Graph neural networks (GCN, GAT, MEGNet, M3GNet) for structure-aware prediction
  • Transformer-based models for multi-property estimation
  • Materials and protein language models (MatBERT, ESM-2)

These components operate within a Pareto-guided inference loop, enabling real-time exploration of vast chemical, structural, and biological design spaces under practical constraints.

2️⃣ Physics-Based Simulation (Materials Layer)

We perform scalable first-principles simulations to validate and refine AI-generated materials:

  • Formation energy and thermodynamic stability
  • Magnetic properties (Ms, MAE, Tc)
  • Electronic structure and band gaps
  • Crystal relaxation and structural optimization

Our workflows leverage:

  • GPAW and ASE for DFT
  • PyXtal for structure generation
  • M3GNet for ML-accelerated relaxation
  • Compatibility with VASP-style workflows

This ensures that all candidates are evaluated under physically consistent constraints.

3️⃣ Bio-Extraction Modeling (Biomining Layer)

TheraJect extends AI-driven discovery into rare-earth extraction technologies.

This layer includes:

  • Lanmodulin-based selective REE binding systems
  • Protein language models (ESM-2) for sequence embedding
  • ML prediction (XGBoost / RF) of binding selectivity and performance
  • Immobilization and support-material optimization

This enables the design of sustainable, low-impact extraction processes as an alternative to conventional chemical methods.

4️⃣ Multi-Objective Optimization & Decision Layer

All candidates—materials and biomining systems—are evaluated through Pareto-based multi-objective optimization, balancing:

  • Performance (magnetics, selectivity, efficiency)
  • Thermodynamic stability and feasibility
  • Cost and supply-chain criticality
  • Manufacturability and scalability

This enables transparent trade-off analysis and risk-aware decision-making.

5️⃣ Data Infrastructure & Data Fusion

The platform integrates diverse datasets, including:

  • Materials Project, OQMD, JARVIS, AFLOW
  • Matminer-derived feature databases
  • LanM and biomining datasets
  • AI-generated synthetic datasets
  • Literature and experimental data

All data streams are normalized and continuously refined through feedback loops between AI models, simulation, and real-world constraints.

6️⃣ Digital Twin & System-Level Modeling

TheraJect extends beyond discovery into system-level optimization through digital twin modeling.

This layer enables:

  • Simulation of rare-earth supply chains
  • Modeling of extraction, separation, and recycling processes
  • Scenario analysis under geopolitical and policy constraints
  • Optimization of cost, resilience, and environmental impact

This transforms isolated discovery into actionable, system-level innovation.

7️⃣ End-to-End Inference Loop

TheraJect’s platform operates as a continuous discovery and optimization loop:

AI-based generation  
→ ML prediction  
→ Pareto optimization  
→ Physics / biological validation  
→ System-level evaluation (digital twin)  
→ Down-selection for deployment

This framework reduces traditional R&D cycles from months to days while improving accuracy, interpretability, and real-world applicability.

Our Approach

TheraJect builds on open scientific methods.

Our innovation lies in integrating AI, physics, biology, and system modeling into a unified platform for real-world rare-earth innovation under physical, economic, and societal constraints.