Planners approve
Nothing runs on Fluxline’s say-so. A planner approves each daily schedule and can move any job.
Manufacturing production planning
Fluxline helps small manufacturers sequence production orders to cut cleaning, tooling and setup time between jobs. Machine learning estimates changeover durations from production records; an optimisation engine recommends schedules that respect deadlines and machine availability — for planners to approve.
In the order received
Sequenced by Fluxline
The problem
Going from a dark ink to a light one means a full wash-up. A new substrate means re-threading; a different width or plate set means re-tooling. The order you run jobs in decides how much of the shift goes on making product — and in most small plants it is decided on a whiteboard.
What drives changeover time on this fictional line, learned from 200 past changeovers. Typical error on 40 held-out changeovers: ±1.9 min.
| 404 | 401 | 405 | 402 | 403 | 409 | |
|---|---|---|---|---|---|---|
| 404 Bread bag, white | — | 10 | 28 | 49 | 47 | 48 |
| 401 Bread bag, wholemeal | 14 | — | 26 | 47 | 45 | 46 |
| 405 Detergent refill | 35 | 29 | — | 41 | 47 | 48 |
| 402 Rice sack 25 kg | 60 | 55 | 45 | — | 49 | 50 |
| 403 Snack pouch, chilli | 62 | 56 | 54 | 53 | — | 21 |
| 409 Coffee pouch | 63 | 57 | 55 | 54 | 21 | — |
Production sequencing
Fluxline searches thousands of possible job orders, scoring each by total changeover time and any lateness — with lateness weighted twenty times heavier. The planner sees the recommendation, the reasoning and the alternative, then approves.
Light to dark, same substrate and plates together, urgent jobs early. Fictional jobs.
Scenario comparison
Add an order or block out the press, and Fluxline re-plans in seconds — showing changeover time, late jobs and when the shift finishes, before anyone commits.
When something has to give, Fluxline says which job will be late and by how much — the planner decides.
Performance tracking
Each shift’s actual run and changeover times are recorded against the plan. The gaps show where estimates are off — and every new record makes the next estimate better.
planned actual
How it works
Upload your job list, changeover log and shift reports as spreadsheets.
Fluxline fits a changeover model to your own records and shows its error.
It recommends an order for the next day that respects due times and downtime.
The planner adjusts if needed and approves; actuals flow back after the shift.
Nothing runs on Fluxline’s say-so. A planner approves each daily schedule and can move any job.
Every changeover estimate is shown with the model’s typical error on records it never saw.
The sequencer would rather spend changeover minutes than miss a due time — and says so when it can’t avoid it.
Estimates learn from your own line, not an industry average.
Why it matters
Share of US manufacturing firms with fewer than 20 employees — plants that usually plan production by hand.
National Association of Manufacturers, Facts About ManufacturingManufacturers in the United States, employing 12.6 million people.
National Association of Manufacturers, Facts About ManufacturingShigeo Shingo, creator of SMED, reported that the setups he worked on fell on average to 2.5% of their original time — changeover time is far from fixed.
Wikipedia, “Single-minute exchange of die”Who it’s for
Films, bags and cartons with frequent size and print changes.
Ink, plate and substrate changes between short runs.
Colour and material purges on extrusion and moulding lines.
Many customers, many recipes, cleaning between each.
Pricing
Illustrative pricing for a proposed product, to be agreed with pilot plants.
$290per line / month
$690per month
Custom
Technology
Fluxline’s work is a cycle: production records come in every shift, changeover models are retrained as more records arrive, and sequencing is a compute-heavy search that runs whenever an order or a machine changes. Designing on AWS lets storage, model training and on-demand optimisation run as managed services, and keeps each manufacturer’s data isolated as they grow from one line to a whole plant.
Amazon S3AWS Lambda
Spreadsheet exports of jobs, changeover logs and shift reports land in per-customer storage; Lambda validates and normalises them.
Amazon SageMaker AI
A duration model is trained on past changeovers — shade, material, width and tooling either side — and reports its error on held-out records.
Amazon DynamoDB
Open orders, due dates, machine calendars, material availability and approved schedules are stored per line.
Amazon ECSAWS Fargate
Containerised optimisation runs on demand to search job sequences that cut changeover time without missing due dates.
Amazon API GatewayAmazon Cognito
Authenticated schedule, scenario and approval APIs serve the schedule board, with planner and supervisor roles.
Across every stage
| Service | Role |
|---|---|
| Amazon S3 | Stores imported job lists, changeover logs, shift reports and exported schedules, separated by customer. |
| AWS Lambda | Validates imports, triggers re-planning when an order or machine changes, and records planned-vs-actual results. |
| Amazon SageMaker AI | Trains and hosts the changeover-duration model for each line, and tracks its accuracy as new records arrive. |
| Amazon DynamoDB | Holds jobs, due dates, machine calendars, constraints, approved schedules and the decision log. |
| Amazon ECS | Runs the sequencing optimiser as a containerised service, isolated from the rest of the platform. |
| AWS Fargate | Supplies compute for each optimisation run on demand, so a long search never slows the workspace. |
| Amazon API Gateway | Exposes the schedule, scenario and approval endpoints, plus the website’s pilot-request intake. |
| Amazon Cognito | Signs users in and separates planner, plant-manager and supervisor roles. |
| Amazon SES | Sends approved schedules and pilot-request notifications. |
| Amazon CloudWatch | Collects logs and metrics for imports, model training, solver run time and API errors. |
| AWS KMS | Encrypts production data at rest with customer-specific keys. |
Proposed, not deployed. This website and the schedule board run in your browser on fictional data.
Team
Founder
Augustina founded Fluxline after watching small plants lose hours a day to changeovers that a better job order would have avoided — planned on a whiteboard because scheduling software was built for factories ten times their size. Augustina leads product, customers and every pilot.
Progress & pilot programme
Fluxline is in its building stage and opening its first pilots. A pilot covers one production line: we import your job lists and changeover records, fit the model, and your planner approves a recommended schedule each day while we compare plan and actual.
| 2026 | Company foundedBy Augustina Oyetade, to give small plants the scheduling tools large ones take for granted. |
|---|---|
| Done | Planning method definedChangeover model from your own records, a sequencer that respects due dates and downtime, planner approval. |
| Done | Schedule board prototypeSequencing, scenarios, approval and planned-vs-actual shown on a fictional flexo line. |
| Done | AWS reference architectureData intake, model training, optimisation and planner workspace designed. |
| Next | First pilotsOne production line per pilot, from your spreadsheet exports, with schedules your planner approves daily. |
FAQ
A list of upcoming jobs with due dates, and a log of past changeovers — even a rough spreadsheet. The more history, the better the estimates.
No. It recommends a sequence and a planner approves it. It does not connect to or control equipment.
The planner can move any job; Fluxline re-calculates changeover time and lateness straight away.
Not yet. It is a proposed product; the schedule board is a prototype running in your browser on fictional jobs.
Contact
Questions about the product, the pilot programme, spreadsheet imports or how production data is handled. Messages go directly to the founder.