Turning 1,200+ operating parameters into a factory control system

The line's operating parameters — the centerlines that define what 'running correctly' actually means — were spread across machines, spreadsheets, and the experience of individual operators. Without one authoritative definition of the controlled state, deviations were caught late, corrective action depended on who was on shift, and automation had nothing stable to build on.

• Consolidated and analysed 1,200+ machine centerlines across the Beauty Packing line into a single controlled-state definition.

• Built and piloted the Controlled State system directly on the floor at P&G's largest global manufacturing site.

• Coordinated Operations, Quality, and Engineering contributors to validate parameters against how the line actually runs.

• Defined the implementation roadmap for downstream error reduction and process automation.

• Separately led the annual Quality risk assessment for mis-pack, mis-code, and mis-label detection sensors, coordinating 20+ stakeholders across four teams.

• Designed, executed, and validated a new SOP extending shelf life for a critical body wash ingredient.

• Built an AI agent to streamline site validation document approvals for the Quality department.

1. 01

   Started on the floor rather than in the data: mapped which parameters genuinely govern quality outcomes versus which were inherited and never revisited.

2. 02

   Normalised centerlines into one structure so a deviation on any machine reads the same way, making the line's state legible at a glance.

3. 03

   Pressure-tested the definition with the people who run the line — Operations, Quality, and Engineering each hold a different piece of the truth.

4. 04

   Piloted on a live line, then wrote the roadmap so the system outlives the internship and gives automation a stable foundation to sit on.

5. 05

   For the validation-approval bottleneck, built an AI agent that triages and pre-checks documents while keeping human approval as the decision point.

• Technical: 1,200+ parameters with real interdependencies, where an over-tight specification stops the line and a loose one ships defects.

• Organisational: three functions with different incentives — Operations optimises throughput, Quality optimises risk, Engineering optimises capability — had to agree on one definition.

• Environmental: a live, high-volume line at the company's largest global site. Nothing could be tested by taking production down.

• Durability: the system had to be operable by people who were not in the room when it was designed.

I go to where the knowledge actually lives — the floor — and turn scattered operating experience into a system other people can run without me. The hard part was never the analysis; it was getting three functions to agree on one definition of correct, then leaving behind something durable enough to automate against.