A prototype that is honest about being a prototype.
WellTwin AI was built to answer problem statement 26120 with a working decision workflow rather than a slide deck. It is a demonstration on synthetic data and it says so on every screen.
What this build does
- Models a full CSS cycle from steam design through soak to production decline.
- Couples that inflow to a rod-pump model so fillage and rod loading respond to reservoir changes.
- Lets an operator test scenarios side by side before committing a cycle.
- Ranks field-wide opportunities under a steam constraint.
- Turns a model output into a guard-railed recommendation with an accept/reject audit trail.
- Exposes data quality, model drift and feature-level explanations alongside every number.
What this build is not
- Not connected to Oil India Limited systems, SCADA, historians or any proprietary dataset.
- Not a calibrated reservoir simulator; the CSS model is a reduced-order surrogate.
- Not validated against measured field response, so no accuracy claim is made.
- Not a control system; nothing here writes a set-point to any equipment.
- Not a security implementation; the sign-in is a mock client-side gate for the demo.
- Not a commercial product or an endorsement by any operator.
How the synthetic data works
A seeded pseudo-random generator derives every well property from a stable string key, so 12 wells, their telemetry traces, cycle histories, alerts and work orders are identical on every machine and every reload. Relationships between quantities are enforced rather than random: water cut climbs through a cycle, steam-oil ratio follows tonnage and response, failure risk follows rod loading and low fillage, and health score is the inverse of risk.
BGW-08 is the scripted demo well. It carries a deliberate, diagnosable fault: production has fallen while the annular fluid level has risen and fillage has decayed to 63%. That signature is a lift problem, not depletion, and the twin, the explainability view and the recommendation all resolve to the same conclusion — which is the point of the walkthrough.
Demo access
On the sign-in page, use the Use demo credentials button — it fills the demo account and signs you in with one click.
Signing in stores a local demo session in your browser and opens the Command Center. No account is created and nothing leaves your device.
Sign in to the demoSuggested walkthrough for evaluators
Roughly five minutes end to end
- 1Command Center — See the field state and notice BGW-08 flagged with a high-severity fillage alert.
- 2Alerts & Risk — Open ALM-1041, read the signature, and acknowledge it.
- 3Well BGW-08 → Digital Twin — Inspect the well-to-surface schematic and modelled vs measured channels.
- 4Model Explainability — Confirm the diagnosis is a lift problem, not reservoir depletion.
- 5What-If Simulator — Lower pump speed, lengthen the stroke, and compare against baseline.
- 6Operator Recommendation — Accept REC-2207 with the guardrails shown, creating an audit entry.
- 7Reports — Generate and download the cycle performance report.
The Judge Demo screen inside the app runs this same path with direct links at each step.
Attribution and responsible-use statement
The Baghewala field and Oil India Limited are referenced only to describe the problem context of Smart India Hackathon problem statement 26120. This prototype is an independent student work, uses no operator data, and carries no endorsement. Nothing shown should be read as a performance claim about any real asset. Any modelled uplift figure in this product is an output of a simplified surrogate on fabricated inputs, and full engineering validation, field calibration and management of change would be required before any of it informed a real operating decision.