Cogentic

How it works

Sheet 03 · Rev A

The foundation first. Then the software.

Get the data out of every system, build the foundation under it, then put software on top that your own people run.

The whole chain

The whole chain, not one link of it.

We are engineers from the buildings we sell into. Both founders work every engagement.

  • Armaan Waels, co-founder of Cogentic. Black-and-white portrait.Armaan Waels, co-founder of Cogentic. Black-and-white portrait.

    Armaan Waels

    Co-Founder · Forward-deployed engineering

    • Payment infrastructure at a Y Combinator-backed fintech.
    • Event-driven pipelines behind 100,000 transactions a day.

    Georgia Tech · CS, Machine Learning

  • Aymane Arfaoui, co-founder of Cogentic. Black-and-white portrait.Aymane Arfaoui, co-founder of Cogentic. Black-and-white portrait.

    Aymane Arfaoui

    Co-Founder · Data platform and ML

    • Three years of ML infrastructure for aerospace manufacturing: millions of telemetry records a day, statistical process control and anomaly detection on live production lines.
    • Patent pending on uncertainty-aware prediction.

    Georgia Tech · CS, Machine Learning

Part 01

Ensure the data exists.

Exports from the ERP and logs straight from the machine. If a record only exists on paper, we digitize it. Wherever it lives, we go and get it.

Sources so far: ERP, MES, PMS, WMS, and EHR exports · machine logs · scanned paper · spreadsheets

  • ERP exportCSV
  • PLC / machine logTEXT
  • Quality systemSQL
  • Inspection formSCANNED PDF
  • SpreadsheetXLSX ×5
  • NotebookPHOTOGRAPHED

Ingest

Any format · anywhere it lives

Part 02

Build the foundation.

Pipelines and a data model built around your operation, down to your part numbers and your exceptions. Three stages, called medallion architecture.

Select a tier

Structured tables built around the questions you ask.

scrap_by_alloy_machine_shift · tool_change_windows · example

Running AutoML on scattered data is a Ferrari with no road.

Most companies sell the car. We build the road first.

Part 03

Software on top.

It runs on a machine inside your building. Offline, air-gap capable. Your people run it, no code and no data scientist, in four steps.

Upload · Profile · Train · Predict

It trains competing models and promotes the winner. The answer comes back ranked, with a confidence score, and every model traces to a hashed, re-runnable plan.

01 · Uploadexample · offline
  • scrap_log_2024.csvok
  • tool_changes.xlsxok
  • shift_schedule.csvok
  • machine_log_q1-q4.txtparsed
4 files · joined on job_no via the Gold layer
The run, end to end · scroll · example data03 / 04
NO FIELD JOINS7 SOURCES · 0 SHARED KEYSBRONZEeverything centralized in one place, as-is01SILVERcleaned, deduplicated, validated, every fix versioned02GOLDstructured tables around the questions you ask03QUERY · TUESDAY, 9:14Do we scrap more on this alloy,machine 4, second shift?RANKED ANSWER · EXAMPLEalloy 7075 · machine 4 · shift 2scrap risk HIGH · confidence 0.87plan sha256 7f3a…c1 · re-runnableCOGENTIC · DEPLOYMENTRuns on a machine insideyour building. Offline.AIR-GAP CAPABLE · 0 OUTBOUND CONNECTIONSUPLOAD · PROFILE · TRAIN · PREDICTERPkey: job_noMACHINE LOGkey: JobQUALITY SYSTEMkey: JobNumberSCANNED FORMkey: none (paper)v3-FINAL-use-this-one.xlsxkey: Job #NOTEBOOKkey: none (handwritten)MAINTENANCE PDFkey: WO

Build the road

Bronze centralizes everything as-is. Silver cleans and versions it. Gold holds the tables your questions run against.

The run, end to end · scroll · example data03 / 04
ERPjob_noMACHINE LOGJobQUALITY SYSTEMJobNumberSCANNED FORMnone (paper)v3-FINAL-use-this-one.xlsxJob #NOTEBOOKnone (handwritten)MAINTENANCE PDFWOBRONZEeverything in one place, as-is01SILVERcleaned, validated, every fix versioned02GOLDtables around the questions you ask03QUERY · TUESDAY, 9:14Do we scrap more on this alloy,machine 4, second shift?RANKED ANSWER · EXAMPLEalloy 7075 · machine 4 · shift 2scrap risk HIGH · confidence 0.87COGENTIC · DEPLOYMENTRuns on a machineinside your building.Offline.AIR-GAP CAPABLE · 0 OUTBOUNDUPLOAD · PROFILE · TRAIN · PREDICTERPkey: job_noMACHINE LOGkey: JobQUALITY SYSTEMkey: JobNumberSCANNED FORMkey: none (paper)v3-FINAL-use-this-one.xlsxkey: Job #NOTEBOOKkey: none (handwritten)MAINTENANCE PDFkey: WO

Build the road

Bronze centralizes everything as-is. Silver cleans and versions it. Gold holds the tables your questions run against.

The run, end to end · example data

Build the road03 / 04
ERPjob_noMACHINE LOGJobQUALITY SYSTEMJobNumberSCANNED FORMnone (paper)v3-FINAL-use-this-one.xlsxJob #NOTEBOOKnone (handwritten)MAINTENANCE PDFWOBRONZEeverything in one place, as-is01SILVERcleaned, validated, every fix versioned02GOLDtables around the questions you ask03QUERY · TUESDAY, 9:14Do we scrap more on this alloy,machine 4, second shift?RANKED ANSWER · EXAMPLEalloy 7075 · machine 4 · shift 2scrap risk HIGH · confidence 0.87plan sha256 7f3a…c1 · re-runnable
Bronze centralizes everything as-is. Silver cleans and versions it. Gold holds the tables your questions run against.
Inside your building04 / 04
COGENTIC · DEPLOYMENTRuns on a machineinside your building.Offline.AIR-GAP CAPABLE0 OUTBOUND CONNECTIONSUPLOAD · PROFILE · TRAIN · PREDICT
It runs on a machine inside your building, offline. Upload, Profile, Train, Predict.

What you own

You own the pipelines.

You own the data model.

You own the structured data.

You own the models.

A consultant

leaves a report.

A data scientist

leaves in 18 months and takes the capability along.

Cogentic

leaves a working function that your own engineer re-runs.

In productionFortune 100 manufacturer · fully air-gappedFirst spreadsheet to validated model · 3 weeksRe-run weekly by their engineer

Next

See it on your own data.

The exploratory phase runs one of your datasets through the pipeline and trains one model on your history. $12,500, 30 to 45 days.

The $12,500 exploratory phaseBook a call