The Technology
The Steel Reasoning Architecture
One physics-grounded multi-agent engine that reasons over real steel-plant physics — the same architecture across the crane, the furnace, the ladle, and the caster. The orchestration is shared; the physics layer is steel-specific.
The Problem
Most steel-floor AI is a point solution, a language copilot, or a manual interpretation loop
Single-asset point tools are useful on one machine but cannot reason across crane, furnace, ladle, and caster. GenAI copilots reason only at the language layer with no steel-physics understanding. On most floors the interpretation still happens in the operator's head — slow, reactive, and hard to explain.
Point Solutions
- Useful on one machine
- Blind beyond that asset
- Manual interpretation loops
GenAI Copilots
- Language-layer reasoning only
- No steel-physics understanding
- No grounding in the floor
Current Operations
- Reactive, not predictive
- Interpretation lives in the operator's head
- Siloed asset data
The GoatAI Difference
Physics-grounded multi-agent orchestration
Four integrated layers. 01 Data Ingestion: furnace & ladle video, crane geometry, radar clearance, level, telemetry, PLC/SCADA, thermal/IR. 02 Physics Integration: BOF kinetics, heat-transfer models, caster solidification, crane kinematics, metallurgical endpoint models. 03 Orchestration: monitoring, prediction, reasoning, and decision agents. 04 Application: pulpit dashboards, clearance & endpoint alerts, visibility walls, routed recommendations.
01 — Data Ingestion
02 — Physics Integration
03 — Multi-Agent Orchestration
click an agent to inspect
04 — Application Layer
→ Clearance alerts · Endpoint & quality reasoning · Pulpit visibility
One architecture across crane, furnace, caster, and ladle
Core Capabilities
Why autonomous physical reasoning is different
Real-Time Physical Reasoning
Multi-agent systems that continuously understand evolving conditions on the floor — crane clearance, furnace state, ladle condition, and caster behaviour — grounded in steel physics.
Operator-Facing Decision Workflows
No manual interpretation loops. Agents monitor, predict, reason, and route recommendations to the pulpit — while deterministic barriers always own the stop.
One Architecture, Steel-Wide
The same reasoning architecture runs the crane, the furnace, the ladle, and the caster. The physics layer is steel-specific; the orchestration is shared.
Results
What physics-grounded agentic AI delivers on the floor
Clearance alerts
Camera and radar fused into clearance reasoning — flagging deviation from the bay's normal state, while a deterministic barrier owns the stop.
Endpoint reasoning
Physics surrogates over the BOF converter and caster support endpoint and quality reasoning.
Pulpit visibility
Low-latency operating-pulpit video over furnace shell and lance-interaction zones, on the visibility wall.
Explainability
Steel physics grounds the reasoning. Operators understand why, not just what.
Steel Proof
Same engine. Three groundings inside steel.
This is not a point solution. The same reasoning architecture runs the crane & yard, the furnace & converter, and the ladle & caster — each with its own steel physics. Built from the plant floor up, against live constraints.
Crane & Yard
Clearance & Position Reasoning
Recognition-based clearance management and load monitoring over overhead cranes and the slab & coil yard.
- HookVision — crane anti-collision, fusing camera + radar (in development)
- Novelty and deviation flagging against the bay's normal state
- YardVision — fixed-optical slab & coil position and load monitoring (in design)
Same reasoning architecture as the furnace and ladle. Here the physics is crane kinematics and clearance geometry — and a deterministic barrier always owns the stop.
Explore Crane & Yard →
Furnace & Converter
Pulpit Visibility & Endpoint Reasoning
Operating-pulpit video and physics surrogates over the EAF and BOF converter for visibility, endpoint, and quality reasoning.
- HeatVue — EAF operating-pulpit video wall, 4×5MP GMSL2, ≤60 ms lens-to-output (Rev F offer)
- Furnace shell and lance-interaction zone coverage
- BOF surrogates — reduced-order models + ML for endpoint & quality reasoning (research)
Same reasoning architecture as the crane and ladle. Here the physics is BOF two-zone kinetics and metallurgical endpoint models.
Explore Furnace & Converter →
Ladle & Caster
Ladle Condition & Solidification Reasoning
Ladle visualization across the DE bay and solidification surrogates over the continuous caster.
- LVS — ladle ID, lining condition, lift/landing safety, slag & skull (in design)
- DE bay coverage: 36 ladle stands, 3 cranes, 15 m hook height
- Caster surrogates — solidification ROM for quality reasoning (research)
Same reasoning architecture as the crane and furnace. Here the physics is heat transfer and caster solidification.
Explore Ladle & Caster →
Watch the agents reason in real-time
Interactive walkthrough of the SENSE → UNDERSTAND → DECIDE → ACT cycle