05A multi-tenant Pharma Manufacturing Analytics SaaS

SPC & Process Capability Analytics

Built and owned the engine

Statistical process control for pharma manufacturing, used by 10+ enterprise clients across their sites. Control charts, run-rule alerts and process capability (Cp/Cpk/Pp/Ppk) come from live batch data, with correct handling of skewed data. I built the first version of the engine and owned it for over three years.

13: chart types (I-MR, EWMA, Pareto…)
10+: enterprise clients, multiple sites each
Cp · Cpk · Pp · Ppk: capability, with Box-Cox or Yeo-Johnson for skewed data

Python / Statistics / PostgreSQL

Problem

Pharma manufacturers must show regulators that each process stays in control across hundreds of batches. Quality engineers needed control charts with limits that change over time, run-rule detection and capability indices, computed from live batch data rather than spreadsheets. Real batch data is often skewed or one-sided, which makes naive Cpk numbers misleading.

Approach

I built the first version of the charting and capability engine and owned it as the platform grew.
One API serves 13 chart types, from I-MR and EWMA to Pareto.
Limits are date-aware, so each batch is judged against the limit that applied at the time.
Run rules are configurable per chart, and every flagged point records why, for the audit trail.
Capability analysis checks normality first and transforms skewed data before it computes the indices.

Architecture

Batch data
Filters & exclusions
Date-effective limits
Run rules
Capability indices
Charts & violations

Outcome

The engine runs across 10+ enterprise pharma clients, each with multiple manufacturing sites. Quality teams get control charts, violation tables and capability scorecards from live batch data, with a full audit trail.

What I learnedA transformed Cpk only means something if the specification limits go through the same transform as the data.

Building something like this?

Tell me about it