Phase 0 · Data Foundation

Data Request

To  IT / BMS Integration
From  —
Return by  —

Exports below are needed before AI model development can begin. Where a full range is unavailable, send what exists and note the earliest reliable date per site rather than withholding the export. Partial data is workable; unknown data coverage is not.

Exports

ID Export Range & grain Required fields
E1 Point / tag manifest Current · all sites pointId, siteId, equipmentId, description, unit, dataType, protocol, expected interval
E2 Interval history · critical 24 months · 15-minute pointId, timestamp (ISO 8601, with timezone), value, quality flag
E3 Alarm history 12 months · per event alarmId, pointId, siteId, severity, text, raisedAt, ackAt, clearedAt, operator
E4 Work order / service history 24 months · per order woId, siteId, equipmentId, raisedAt, closedAt, fault description, parts replaced, engineer
E5 Equipment register Current · all sites equipmentId, type, make, model, capacity, design conditions, install date, parent system
E6 Meter register Current · all sites meterId, siteId, role (main / DG / solar / sub / reference), CT ratio, last calibration date
E7 DISCOM bills 24 months · monthly Original PDFs per site — needed to validate the tariff model against real billed amounts

Access required

  • Read-only historian account, all sites
  • VPN or jump-host route to the BMS network
  • Non-production sandbox mirroring one live site
  • Read access to the alarm and work-order systems

Delivery

  • Parquet preferred for E2, CSV (UTF-8) acceptable
  • One file per site per export, named E2_<siteId>_<from>_<to>
  • All timestamps in IST with explicit offset
  • Transfer via SFTP or encrypted drive, not email

Questions to answer in writing

These are not exports. Each one changes how the data is interpreted, and a wrong assumption here invalidates a model rather than delaying it.

  1. Earliest date each site's interval data can be relied on — not when logging started, but when it became trustworthy.
  2. Which sites have a calibrated reference meter, and when each was last calibrated.
  3. Known outage, migration or data-loss windows to exclude from training.
  4. Whether documented comfort and safety limits exist per system, or whether they live with individual engineers.
  5. Any point whose tag name does not match what it physically measures — known mislabelling is cheaper to declare than to discover.
  6. Whether client contracts permit model training on tenant data, and whether that data may leave the country.
If the full set is slow to assemble, send E1 and E2 for one site first. A single site with twelve months of interval data is enough to begin, and it tells us within a week whether the acceptance criteria in the model brief are achievable on real data.
BuildOptix Solutions · Data Request · Phase 0 Queries to: —