Back to PRAVAH

About / Method

PRAVAH reads a spreadsheet and says two separate things: whether the table is clean, and whether it is ready to be treated as a process. It does not call a language model. Personal columns are hashed before any scoring, and the upload is not written to disk.

What it does now

Data quality runs on every table: missing cells, duplicate rows, mixed types, and outliers. Process readiness runs only when a case, an activity, and a time are all detected. A check that does not apply is marked “Not applicable” and does not reduce a score.

The sample-size line is the usual margin-of-error formula, n = 1.96² × 0.5 × 0.5 / 0.10² = 97 cases. Fewer cases are shown as a preview.

Pipeline

Each box is a real step. The page ticks a step only after that step has finished.

  1. 1. Reading fileCSV, Excel, or XES, from memory. Limit 3 MB.
  2. 2. Resolving datesYear-first and ISO dates skip the day/month guess. One decision is shared across a wide sheet.
  3. 3. Hashing personal dataNames, emails, phones, and profile URLs, in any column position.
  4. 4. Detecting shapeEvent log, wide sheet, or a plain table. A plain table is not forced into a process.
  5. 5. Checking qualityThe data-quality score, on the sheet before a wide sheet is melted.
  6. 6. Building readinessOnly when case, activity, and time exist. Otherwise the score stays blank.

What is not built yet

Bottlenecks, survival curves, open-case risk, cause checks, what-if, and the action plan are computed from the uploaded table. Process replay and funnels wait until that map exists. The page does not animate a result it has not computed.

Author

Designed & built by Koushal Acharya

Let's connect with Koushal

🔒 Uploads are processed in memory and never stored.