How we work
Nothing gets quoted before someone has been to the site
The same sequence whether it is one inspection cell or a multi-site program. What changes is the number of phases, not the discipline.
How an engagement runs
- Step 01
Look before quoting
A four-week AI Pilot at a fixed price: workshops with the people doing the work, an inventory of the data or footage you hold, and working software against a real slice of it. For vision work this includes a site survey.
- Step 02
Agree what working means
Accuracy, latency, throughput and uptime written as numbers, with the method for measuring each. Scope is frozen against those figures before build starts.
- Step 03
Build in phases
Each phase ends in something you can use, deployed in your environment and reviewed with your team. Written weekly reporting on scope, progress and risk throughout.
- Step 04
Commission and prove
Hardware installed and integrated where relevant, then a supervised run in shadow mode. The result against the agreed thresholds is published whether it passes or not.
- Step 05
Hand over or operate
Runbooks, training and full assignment of source, weights and datasets — or we stay on under an operations agreement.
Pilot plan
What four weeks actually looks like
The AI Pilot, week by week. Fixed price, and the acceptance criteria are signed before week two starts.
- Week 1
Scope and sign
Workshops, data review, and a site survey where the work is physical. Acceptance criteria agreed in writing before anything is built.
- Week 2–3
Build
Working software against a real slice of the problem, on your data, reviewed with your team mid-way.
- Week 4
Evaluate
Tested against the criteria from week one, with the method and the held-out set documented.
- Readout
The recommendation
The result, a costed path to production, and a clear call. Including do-not-proceed, which we have given before.
Visibility
Written reporting, not a status meeting
Every week: what changed, what it measured, and the decision we need from you.
- • Work completed and work opened, with links to the change
- • Current measured accuracy against the agreed threshold
- • Deployments, commissioning steps and anything reverted
- • Blockers, and the decision we need from you to clear them
# PranInnovations · weekly progressphase build (2 of 4) · cell-01 inspectiontarget >= 99.0% detection, <= 20 ms decisionmeasured 99.4% detection · 12 ms decisionopen tool-witness class below thresholdcommissioning camera + enclosure mounted, PLC pendingnext bench test OPC UA write pathblocker need change window for controls team
Start with a call, then a costed plan
Thirty minutes on the problem, the site and the constraints. If it looks like a fit, the next step is a four-week AI Pilot at a fixed price, with acceptance criteria signed before any code is written.