Computational Instrumentation
Learn how your process behaves. Then control it with confidence.
We turn sensor data into models that simulate, tune, and control real manufacturing equipment.
Founder
AI/ML Researcher — Computational Instrumentation
Vivek works at the intersection of engineering and machine learning — modeling physical processes, then using that to control them.
His process experience includes optical fiber manufacturing, one of the hardest physical processes to instrument and control, in collaboration with MIT's Device Realization Lab under Dr. Brian W. Anthony.
Proof Point
Case Study — Optical Fiber Drawing Control
Fiber drawing is hard to control — the physics are nonlinear and delayed, so a fixed model doesn't hold up. Instead, an LSTM network learned the plant's real dynamics from production data, and system identification modeled the controllers under real noise and disturbances. Together, they formed a closed-loop simulation — a digital twin that replaces slow, trial-and-error tuning on live equipment.
Separately, deep reinforcement learning controlled fiber diameter directly, outperforming standard PI and QDMC control on changing target trajectories.
Two approaches, one idea: learn the dynamics, then simulate and control.
Proof Point — Digital Systems
Case Study — Security Telemetry Health Monitoring
A "sensor" doesn't have to be hardware. A security tool — an endpoint agent, firewall, or log forwarder — reports in at expected intervals, and can go silent the same way a physical sensor can. Statistical baselining catches thousands of these feeds going quiet or degrading, before they become blind spots, running as a multi-tenant pipeline with an agent engineers can ask about any sensor's health directly.
Same sensor-health thinking. Physical hardware or digital telemetry.
Process
Use your existing sensors and machines. No overhaul needed.
Train a model of how your process actually behaves, from real data.
Test control changes on the model first — not on live equipment.
Roll out the tuned controller, with a safe fallback built in.
Long-term: support safe, constrained intelligent process control.
Vision
The long-term vision: a system that learns how physical processes behave, and helps engineers control them safely and confidently.
This is a roadmap, not a claim of current autonomous control.