System
Graph first, execution second.
The model represents teams, roles, workers, processes, tasks, KPIs, and tools as connected nodes. Simulation and optimization sit beside workflow execution, confidence, approval gates, and a Reality dashboard.
Lab preview · operating systems
SimTwin
A visual business twin for modelling organizations, simulating capacity, and running human and AI workflows with approval boundaries.
What we’re trying to solve
See how work moves before automating it.
Organizations are usually described in documents, while the work happens across people, tools, processes, and handoffs. SimTwin explores a navigable model where capacity, ownership, throughput, and execution can be reasoned about together.
High-level workflow
org graph → capacity / cost / ownership model
↘ simulate → approve → execute → reality metricsEvidence
What the repository makes visible.
| Evidence | What the repository shows |
|---|---|
| Domain model | Graph nodes cover organizational structure, roles, workers, processes, tasks, KPIs, and tools. |
| Execution layer | Phase 11 documents worker registration, Run Now, per-step confidence, approval gates, and execution metrics. |
| Runtime options | A local agent runtime supports offline stubs as well as configurable model providers. |
| Implementation | React Flow, FastAPI, SQLAlchemy, NetworkX, and SQLite/Postgres are documented in the architecture. |
What we’re learning
Automation starts with a better map.
01 · Structure is not workflow
A company chart cannot explain the actual handoffs, queues, and decision points that make work slow.
02 · Simulation needs ownership
Capacity and cost only become useful when each activity has a clear owner and measurable outcome.
03 · Approval is a system primitive
Human judgment should be represented in the workflow rather than added after an autonomous action.
04 · Reality must close the loop
Execution metrics are needed to compare the model with what the organization actually did.
Next
What comes next.
The next milestone is a narrow, reproducible workflow scenario that can compare the simulated path, approved run, and observed result without exposing internal operating data.