I build the operating layer where humans and agents collaborate on industrial data — SCADA, PLCs, OPC UA, ERP, MES, MRP, QMS, WMS, PLM — inside the plant safety boundary. Controls and enterprise systems still speak different languages. Most days the translation is a person copying a number, then joining a call to explain why it changed.
Transient Labs — industrial decision AI, energy systems, portable edge inference, and agents whose evidence can be audited near real equipment.
| Project | Demonstrates | Boundary |
|---|---|---|
| Forge · live | Human-agent collaboration, machine identities, plant telemetry, policy proposals, authenticated approvals | Software-in-the-loop; no commissioned PLC control |
| Transient Energy · live | ISO 50001 EnMS, simulator-backed control path, 49-test browser harness | Synthetic plant scenario |
| Pump edge-twin lab · dataset · model | 40k-row clean-room twin, PyTorch → ONNX/TensorFlow/LiteRT parity, browser ONNX/WASM inference | Read-only PLC/SIS boundary; synthetic data |
| Electrolyser baseline · replay scenarios | Anomaly baselines, replayable telemetry, versioned agent/control traces | Methodology proof, not plant savings evidence |
| Carbon · npm | Agent and MCP toolkit for auditable Scope 1/2/3 workflows | Workflow infrastructure, not finance-grade LCA |
| IBM 150-variable Ising study · results · lab | 15 jobs, 599 quantum seconds on IBM Heron, depth/noise evidence, mitigation ablation, exact classical baseline | No quantum-advantage claim; classical HiGHS wins this instance |
Render demos sleep when idle; first load can take a minute.
Live GitHub queries, not hand-maintained numbers. Repositories with merged work:
| Domain | Upstream |
|---|---|
| Industrial protocols, SCADA, MES | pymodbus · opcua-asyncio · FUXA · OpenMes · thin-edge.io |
| Energy and climate | FlexMeasures · Electricity Maps · CodeCarbon · Home Assistant |
| Data infrastructure | clickhouse-connect · dbt-clickhouse · mcp-clickhouse · clickhouse-go · influxdb-client-python · kagglehub |
| Formal methods, quantum, ML | Lean 4 · physlib · Mitiq · Cirq · TensorFlow |
| Platform, tooling | Svelte · coreutils · LeaderWorkerSet · quicklink · docker-library/docs |
Meters are disputed. Spreadsheets outlive their authors. Finance, controls and operations rarely agree on one baseline. This is an attempt to build the factory operating system.
The path stays classical and bounded: replay first, verify invariants independently, keep safety in the PLC/SIS, require an operator before any consequential action. ONNX, TensorFlow, PyTorch, CUDA and quantum accelerators are implementation details behind that.
Hugging Face · Kaggle · LinkedIn · X · Transient Labs