KAIM / SYSTEM 01 · PRODUCTION / CLIENT NDA
Manufaktura
A production MES/APS for planning, execution, warehouse operations, ERP integration and machine-state telemetry. I own the application architecture and full-stack delivery across .NET 10, React and PostgreSQL under a client NDA.
KAIM / BUSINESS VALUE
Manufaktura connects planning, execution, warehouse operations, ERP and machine telemetry. The goal is not another dashboard. The goal is a shorter path from a shop-floor event to a decision, less manual state reconciliation and traceability that remains attached to the product.
PLC telemetry read cycle
machine state can reach the operational layer while there is still time to act on it
machines represented in telemetry flows
audited telemetry scope, not a claim about total plant equipment
business roles in the RBAC model
from planning and production to warehouse, maintenance, purchasing and customer service
WHAT CHANGES OPERATIONALLY
The highest-value moments are the ones where plan, material, machine and operator must agree on current state. Manufaktura connects those points in one flow instead of pushing the reconciliation cost onto users.
Problems surface while there is still time to act.
Telemetry, performance deviations and planning-risk signals return to the decision layer. A manager does not have to wait for shift close to see that reality is starting to diverge from plan.
Planning and material flow operate against the same constraints.
Machine availability, buffers, stock, orders and ERP data are part of one model. Fewer decisions depend on manually reconstructing state from separate spreadsheets, screens and cross-team conversations.
Quality, batch genealogy and decisions leave an inspectable trail.
Hold and Release gates, product genealogy, RBAC and session audit make it possible to reconstruct not only what happened, but who made a decision and where it happened in the process.
KAIM / TRUST
Scope of responsibility, without inference.
- STATUS
- PRODUCTION / CLIENT NDA
- ROLE
- Sole technical owner / Full-Stack .NET Engineer
- PERIOD
- Sep 2025–present
- OWNERSHIP
- Application architecture, .NET / React implementation, data model, application-side machine contracts, security, observability, automated tests, deployment guidance, UAT and handover.
- NOT OWNED
- PLC firmware and physical machine-control logic. I defined and consumed the application-side contracts rather than claiming ownership of controller firmware.
- LIMITATIONS
- Client and domain data remain private. The public material uses aggregate repository counts, anonymized architecture and measured workflow-level performance evidence.
- VERIFICATION
- Static code, AST and repository audit · 17 Aug 2026; profiled workflow evidence · Aug 2026
KAIM / EVIDENCE
Every claim carries an epistemic status.
Numbers are not treated as equivalent. COUNTED means counted in an artifact or repository, MEASURED means measured in a specific workflow, ESTIMATED is approximate, and DOCUMENTED marks a qualitative workflow or architecture contract.
COUNTED 38 functional domains
ProvenanceStatic repository audit · src/features · 17 Aug 2026
COUNTED 396 active REST endpoints
ProvenanceController and HTTP-contract audit · 17 Aug 2026
COUNTED 1,789 automated test methods
ProvenancexUnit + Vitest suite audit · 17 Aug 2026
Scope955 backend + 834 frontend test methods.
COUNTED 112 persistence entities
ProvenancePersistence entity static audit · 17 Aug 2026
DOCUMENTED 54+ machines represented in telemetry flows
ProvenanceMachine telemetry architecture audit · 17 Aug 2026
ScopeTelemetry scope; not a claim about total factory equipment.
MEASURED ~80% backend resource reduction
ProvenanceProfiled production workflows
ScopeMeasured in selected profiled workflows; not a system-wide claim.
KAIM / SYSTEM MAP
The architecture in one pass.
- 01React / TypeScript operational UI
- 02ASP.NET Core / .NET 10 API
- 03Domain + application services
- 04EF Core / PostgreSQL persistence
- 05ERP + warehouse integration
- 06Machine telemetry + SignalR
- 07Automated tests / UAT / delivery
System flows / 04
How a shop-floor signal becomes an operational decision.
An anonymized cut through three overlapping flows: machine telemetry is normalized and pushed live, planning validates route constraints, and operator plus inventory actions converge on one production-state model.
Shop-floor frames represent pulse, speed, state and device availability.
- 1–3 s polling
- raw telemetry
- fail-soft
A background worker mirrors device state, rejects missing or implausible frames and hands normalized values forward.
- polling worker
- sanity guards
- availability
The diagnostics layer derives production delta, speed and anomalies before a meaningful state change is persisted.
- normalization
- anomaly detection
- machine state
Meaningful state changes are persisted and pushed to clients without forcing a manual refresh of the operational surface.
- SignalR
- selective push
- live state
Dashboard and planning views receive current state, deviations and contract-risk signals.
- dashboard
- alerts
- selective render
A schedule change is checked against machine topology, intermediate buffers, material availability and deadlines.
- routes
- buffers
- materials
Production, downtime and quality reports return to the same domain model and event path.
- operator context
- production report
- downtime / QA
Stock synchronization feeds reservations and warehouse movements without making the external ERP the UI source of truth.
- stock sync
- reservations
- warehouse docs
Flow point Flow point
KAIM / ENGINEERING
What mattered technically.
Planning, execution, stock and machine state had to agree on one operational reality.
Manufaktura connects production intent with execution across planning, machine assignment, warehouse movements, ERP data and operator-facing workflows. The hard part is consistency across boundaries rather than any single screen.
A domain-heavy .NET core with a React operational surface.
The system uses .NET 10, EF Core and PostgreSQL behind a React / TypeScript frontend. The audited surface includes 38 functional domains, 396 active REST endpoints and 112 persistence entities, with machine-state ingestion and SignalR updates forming part of the operational path.
Regression evidence is attached to a system that is actually operated.
The audited suite contains 1,789 automated test methods across backend xUnit and frontend Vitest. Production profiling, UAT and delivery evidence complement repository counts; performance claims remain explicitly scoped to measured workflows.