ForceX AI — AI-for-Energy Platform¶
Status: Active Development | GeoForce v1.1 Deployed Website: forcex-ai.com Platform: platform.forcex-ai.com
Executive Summary¶
Indonesia's first AI-for-energy platform — physics-informed AI agents for nuclear reactors, geothermal reservoirs, oil & gas operations, power grids, and renewable energy. Built for Southeast Asia's energy transition.
ForceX AI combines traditional physics simulation engines (OpenMC, TOUGH2, OpenFOAM) with physics-informed machine learning (PINNs, GNNs, FNOs) and LLM-powered agent orchestration via LangGraph.
Core value proposition: Collapse hours of physics simulation into seconds while maintaining the accuracy and auditability that regulators require.
Products (12)¶
| Product | Domain | Type | What It Does | Status |
|---|---|---|---|---|
| GeoForce | Geothermal | Surrogate | CNN reservoir surrogate (57K params, R²=0.997) | Deployed v1.1 |
| GridForce | Power Systems | Data-Driven | LSTM load forecasting + PPO RL microgrid dispatch | MVP |
| StorageForce | Energy Storage | Data-Driven | PPO RL battery dispatch + degradation prediction | MVP |
| PipeGuard | Oil & Gas | Data-Driven | Autoencoder anomaly detection + corrosion physics | MVP |
| WindForce | Renewables | Data-Driven | PINN wake model + PPO layout optimization | MVP |
| DrillAgent | Oil & Gas | Data-Driven | LSTM formation prediction + ROP optimization | MVP |
| SubsurfaceAI | Oil & Gas | Data-Driven | CNN seismic fault detection + horizon tracking | MVP |
| NeutronX | Nuclear | Surrogate | UNet neutron flux surrogate | MVP |
| SimForce | Nuclear | Agent | LangGraph + real OpenMC simulation orchestration | MVP |
| ReactorMind | Nuclear | Surrogate | GNN digital twin + anomaly detection | MVP |
| PhysicsForce | General | Solver | PINN multi-PDE solver (self-supervised) | MVP |
| PlasmaForce | Fusion | Surrogate | NN scaling-law validator for plasma confinement | MVP |
All 12 products have working MVPs with CLI, FastAPI API, 484 passing tests, and Docker support.
Live Deployments¶
| App | URL | Description |
|---|---|---|
| Landing Page | forcex-ai.com | Company landing page (Cloudflare + VPS) |
| Product Platform | platform.forcex-ai.com | Unified dashboard for all 12 products |
| GeoForce API | platform.forcex-ai.com/api/v1/geoforce/ |
Real model API (Pro tier) |
| HuggingFace | GeoForce-CNN-v1.1 | Published model + dataset |
Technology Stack¶
AI & Machine Learning¶
| Category | Technologies |
|---|---|
| Physics-Informed ML | PINNs (Physics-Informed Neural Networks), GNNs (Graph Neural Networks), FNOs (Fourier Neural Operators) |
| Deep Learning | CNN surrogates, UNet, LSTM, Autoencoders |
| Reinforcement Learning | PPO (Proximal Policy Optimization) for dispatch and optimization |
| Agent Orchestration | LangGraph for multi-step simulation workflows |
| Simulation Engines | OpenMC (nuclear), TOUGH2 (geothermal), OpenFOAM (CFD) |
| Frameworks | PyTorch, scikit-learn, DEAP (genetic algorithms) |
Platform Architecture¶
| Layer | Technology |
|---|---|
| Frontend (Landing) | React + Vite + TypeScript |
| Frontend (Platform) | React + Vite + TypeScript |
| Backend | FastAPI, Python 3.11+ |
| Database | PostgreSQL, SQLite |
| Auth | Firebase Authentication |
| Deployment | Docker, Docker Compose, VPS |
| CDN | Cloudflare |
| Model Registry | HuggingFace Hub |
| Testing | pytest (484 passing tests) |
Multi-Agent Architecture¶
User Query
│
▼
┌─────────────────────────────────────────┐
│ LangGraph Orchestrator │
│ ┌─────────┐ ┌─────────┐ ┌────────┐ │
│ │ Physics │ │ ML │ │ Data │ │
│ │ Agent │ │ Agent │ │ Agent │ │
│ └────┬────┘ └────┬────┘ └───┬────┘ │
│ │ │ │ │
│ ┌────┴────┐ ┌────┴────┐ ┌──┴────┐ │
│ │OpenMC/ │ │CNN/PINN/│ │ Data │ │
│ │TOUGH2 │ │GNN/LSTM │ │ Store │ │
│ └─────────┘ └─────────┘ └───────┘ │
└─────────────────────────────────────────┘
Key Technical Achievements¶
- GeoForce CNN: 57K parameters, R²=0.997 accuracy on reservoir simulation
- 12 Product MVPs: All with CLI + API + tests + Docker
- 484 passing tests across the entire platform
- Physics-informed approach: Models respect physical laws (conservation, boundary conditions)
- Published on HuggingFace: Model weights and geothermal dataset publicly available
- Production deployment: Landing page + platform running on VPS with Cloudflare CDN
Repository Structure¶
ForceX-AI/
├── pages/ # Landing page (React + Vite + TypeScript)
├── platform/ # Unified product portal (React + Vite + TypeScript)
├── products/ # 12 AI products
│ ├── geoforce/ # Geothermal CNN surrogate
│ ├── gridforce/ # Power grid LSTM + RL
│ ├── storageforce/ # Battery dispatch RL
│ ├── pipeguard/ # Anomaly detection
│ ├── windforce/ # Wind PINN + optimization
│ ├── drillagent/ # Drilling LSTM + ROP
│ ├── subsurfaceai/ # Seismic CNN
│ ├── neutronx/ # Nuclear UNet
│ ├── simforce/ # LangGraph + OpenMC
│ ├── reactormind/ # GNN digital twin
│ ├── physicsforce/ # PINN multi-PDE solver
│ └── plasmaforce/ # Plasma NN validator
├── mvp/ # Shared MVP infrastructure
├── scripts/ # Deployment and utility scripts
├── docs/ # Documentation
└── tests/ # Test suite
Skills Demonstrated¶
- Physics-Informed Machine Learning (PINNs, GNNs, FNOs)
- Deep Learning (CNN, UNet, LSTM, Autoencoders)
- Reinforcement Learning (PPO)
- Multi-Agent Orchestration (LangGraph)
- Full-Stack Development (React + FastAPI)
- Model Deployment & Publishing (HuggingFace, Docker)
- Domain Expertise (Energy sector: nuclear, geothermal, oil & gas, renewables)