WITH CLAUDE logo WITH CLAUDE Search

Headroom MP

Assembles Madhya Pradesh’s published substation-loading records into one view of night-time grid headroom for siting data centres, with a cited verdict per factor.

The problem

Madhya Pradesh is courting AI data centres — a 1 GW MoU is signed, one 80 MW site exists in Indore. An AI data centre is a flat 24×7 load. The MP grid is anything but flat: it peaked at 19,902 MW in January 2026 on rabi irrigation and fell to 13,818 MW by July. Solar covers the middle of the day and nothing after sunset, so the binding constraint is a winter night.

Three groups feel this, differently:

  • MPSEDC, answering a hyperscaler asking "can I put 100 MW near Bhopal?" — the honest answer today requires reading transmission PDFs, regulator orders and tender notices by hand.
  • MPPTCL planners, deciding where the next transformer or battery earns its keep, with no assembled view of where load is actually tightest at night.
  • Citizens, indirectly and last — feeder reliability for homes and farms sits in a system that is not public at all.

The problem underneath all three turned out not to be missing data. MPPTCL publishes per-substation loading every month — installed capacity, peak, and the hour the peak was recorded — for 432 substations, going back 55 months. Nobody had assembled it.

How it works

Headroom MP turns that published record into one view that answers the 2 a.m. question with citations.

Three things it does:

  • Reads the government's own documents — Claude parses the spreadsheets and PDFs, merged cells and multi-row headers included, into records that each keep their source, page and month.
  • Shows the network under real load conditions — 432 substations coloured by risk from measured night utilisation, with a worklist of what to act on and who owns it.
  • Scores siting across twelve factors — each a verdict with an argument and a source, rolled up as a tally, never a single number.

And one thing it refuses to do: where the data does not support a verdict, it gives none and names who holds the missing figure. That refusal is the design, not a limitation — a tool that guesses confidently in front of grid engineers is worse than no tool.

Concretely: Birsinghpur runs at 88% of installed capacity on a winter night, 20 MVA spare. Indore runs at 46%, 683 MVA spare. That difference is the whole siting question, and it was already in the public record.

How Claude was used

In the team’s own words, from their submission.

Reads the government's own documents — Claude parses the spreadsheets and PDFs, merged cells and multi-row headers included, into records that each keep their source, page and month.

Team

Tathya

Individual builders are not listed yet — names are published only with their permission.

← Back to the Project Directory