WhiteHelmet

Context

On a large construction project, progress data lives in dozens of disconnected subcontractor spreadsheets — every one with its own columns, formats, and conventions. Directors of operations describe wasting hours chasing files from every system and fixing them by hand just to get one clear picture. The product, Salama, exists to collapse that chaos into a single source of truth.

The recurring complaint across the people we interviewed — finance directors, project directors, and the PMs in the middle — wasn't missing data. It was reconciliation: no single view, stale reports, and hours lost to inconsistent column naming that quietly breaks every aggregation.

My Role

I owned AI consolidation and master-template generation — the two pieces that turn raw, mismatched submissions into a governed output.

For master-template generation, the system supports two paths: a PM can ideate a template conversationally — defining structure, required fields, and validation rules in plain language until the draft matches their needs — or upload an existing Excel template that the system parses and converts into a managed master sheet, preserving the original structure while making it editable and version-controlled.

For consolidation, the key design decision was treating the language model as a compiler, not a data editor. PMs describe what they want in natural language; the model translates that intent into structured operations, which a deterministic spreadsheet engine then executes. The AI never touches the source files directly — so the output is reliable, auditable, and reproducible rather than a black-box rewrite.

How Consolidation Works

One click ingests every subcontractor submission into a single master sheet, normalizing varying formats into a unified output while keeping all logic configurable and PM-controlled.

1. 01SubmissionsHeterogeneous sheets from many subcontractors — inconsistent columns, formats, and schemas, each ingested as a structured snapshot.
2. 02Intent → OperationsPlain-language commands are parsed into structured operations. The AI acts as a compiler, mapping intent to functions — it never edits the source files.
3. 03PM ReviewA deterministic spreadsheet engine executes the operations. PMs review only flagged conflicts; commands apply iteratively until the output is clean.
4. 04Master DatasetA unified, structured master sheet built from confirmed transformations — exportable and ready for leadership, lenders, and auditors.

On top of consolidation sits a reusable formula library — formulas authored in plain language, applied to any column, row, cell, or range, and saved to reuse across templates and reporting cycles. The whole lifecycle, from template creation to final export, lives in one dashboard.

What's Next

The MVP's next steps are integrating with WhiteHelmet's own authentication and database, expanding the generative feature set beyond consolidation and formulas, and tightening the UI for the day-to-day PM workflow. The throughline stays the same: take the interpretation burden off the human and hand back a clean, trustworthy master record.