I run Transient Labs. Public work is industrial AI, energy systems, and auditable agents near plant systems — SCADA/PLC/OPC UA ↔ ERP/MES/PLM.

Transient Energy — ISO 50001 EnMS prototype: simulator-backed control path, 49 browser tests. Live: EMS. Some sections protected under NDA.

Forge — system-of-action demo. Live: forgeagent.onrender.com.

Carbon — agent + MCP toolkit for Scope 1/2/3 accounting. @sankalpsthakur/carbon (0.2.0).

Scope 3 Calculation — local MVP behind that. Baseline inventory, not finance-grade LCA.

• Scope 3 Strategy — CSRD double-materiality / LCA → ESRS
• PlantOpsBench — stub Inspect AI evals for plant-ops decisions
• Plugins — Claude plugins for Scope 3 + delivery

A poker sim note on X (predictable strategies lose under imperfect info) grew into games — 18 game-theoretic environments, same idea, wider harness.

I've been a contributor to pymodbus, opcua-asyncio, FlexMeasures, and electricitymaps-contrib.

Energy flexibility and intervention choice are binary optimization problems — MIP in production today. The same QUBO can run through Qiskit QAOA on IBM Runtime for a side-by-side comparison. Classical stays the production path; quantum is just another solver behind the same interface. I'm exploring that bridge, not claiming advantage.

Experiments I'm interested in:

• Small FlexMeasures-style storage schedule as QUBO; Gurobi/CBC vs QAOA (objective gap + wall time)
• Transient Energy opportunity portfolio under budget/interlocks — MIP vs QAOA, same harness
• Carbon-intensity-aware load shift (Electricity Maps–style signals) as discount/shift QUBO
• Keep PlantOpsBench / agent / OT I/O paths classical; only swap the optimizer when the decision is combinatorial

X · LinkedIn · Kaggle · Transient Labs