The Engineering Underneath the Products
Every capability on this page is running in production at a customer site. Where a technology is not the right answer for your case, we will say so.
22 capabilities across four disciplines
Models that earn their place on the line
We only ship a model when it beats the manual process on your own data. Every capability below is running at a customer site today.
Computer Vision
Defect detection, surface grading and count verification from fixed cameras on the line — the same score every shift, without inspector fatigue.
- Hide and panel grading with an auditable confidence score
- Surface defect classes trained on your own reject library
- Runs on an edge box so the line does not wait on the network
Predictive Analytics
Forecasts that size purchase and production ahead of demand, built on your order history rather than an industry average.
- Seasonal and order-driven demand models
- Reorder points that move with lead-time drift
- Machine baselines that flag wear before a stoppage
AI Agents
Task-scoped assistants wired into the ERP with real permissions — they answer from your ledger, not from a generic corpus.
- Natural-language queries over live ERP data
- Scoped to the user's existing role and row-level rights
- Every action written to the same audit trail as a human edit
Generative AI
Drafting and summarisation where it saves real minutes — quotations, specification notes, audit narratives and customer correspondence.
- Retrieval grounded in your documents, with citations
- Private deployment options where data cannot leave the estate
- Human approval step before anything reaches a customer
Document Intelligence
OCR and extraction pipelines that read invoices, packing lists and test reports straight into the ERP instead of a data-entry queue.
- Layout-aware extraction for messy scans and photographs
- Confidence thresholds route only uncertain fields to a human
- Writes back into the ledger with the source image attached
Instrumentation that survives a plant floor
Industrial networks drop, gateways lose power and sensors drift. The stack below assumes all three and keeps recording anyway.
IoT Platform
Device registry, telemetry store, alert rules and dashboards in one place — the layer everything else on this list plugs into.
- Per-device provisioning, keys and health status
- Threshold, rate-of-change and schedule-based alert rules
- Telemetry retained at full resolution, rolled up for history
Sensor Integration
Temperature, humidity, pH, load, flow, energy and vibration — wired to the process points that actually change a decision.
- Calibration offsets and drift tracking per device
- Analogue and digital inputs via industrial I/O modules
- Placement surveyed on site before anything is quoted
MQTT
Lightweight publish-subscribe transport for constrained links, with QoS and retained state so a reconnect does not lose the last reading.
- TLS with per-device credentials, not a shared password
- QoS 1 delivery with broker-side persistence
- Last-will messages so a dead gateway is visible immediately
Modbus
Modbus RTU and TCP polling for the PLCs, drives, meters and weighbridges already on your floor — no rip-and-replace.
- Register maps built from the vendor manual and verified live
- Serial and Ethernet gateways for mixed-vintage equipment
- Poll rates tuned so old controllers are not overwhelmed
Edge Computing
Rules, buffering and inference on a local box, so control decisions do not depend on a round trip to the cloud.
- Store-and-forward buffering through network outages
- Local rule evaluation with sub-second alerting
- On-device inference for vision and anomaly detection
Device Management
Fleet visibility once the pilot becomes a hundred nodes — firmware, certificates, connectivity and battery in one view.
- Staged over-the-air firmware rollout with rollback
- Certificate rotation and revocation per device
- Health scoring that surfaces the failing node before it is silent
Systems built to be maintained, not just delivered
Most of a system's cost arrives after go-live. We write for the team that inherits it — typed, tested and documented.
Web Applications
Server-rendered business applications that stay fast on a factory-office connection and work on the hardware you already have.
- TypeScript end to end, no untyped boundaries
- Accessible to WCAG 2.2 AA as a build requirement
- Tested against the browsers your staff actually run
Mobile Applications
Android and iOS apps for the people away from a desk — stores, dispatch, field service and management approvals.
- Offline-first data capture that syncs when signal returns
- Barcode, QR and label printing from the handset
- Push approvals so a hold does not wait for someone at a PC
Cloud Applications
Multi-tenant, horizontally scalable deployments for when one plant becomes six and IT does not grow with it.
- Containerised services with health checks and auto-restart
- Blue-green releases so an upgrade is not an outage
- Backups and restores rehearsed, not assumed
API & Integration
The connective work that makes an ERP useful — banks, e-way bills, e-invoicing, couriers, marketplaces and legacy systems.
- Versioned REST endpoints with an OpenAPI contract
- Idempotent webhooks with replay for failed deliveries
- Adapters for legacy systems that have no API at all
Enterprise Architecture
Deciding what to build, what to buy and what to retire — with a migration path that does not require a big-bang cutover.
- Domain modelling before a schema is written
- Clean separation of domain, application and infrastructure
- Phased migration that runs old and new side by side
Where the ledger actually lives
Platform choice follows the workload and your team's skills — we run all of the below in production and will tell you when the cheaper option is the right one.
Azure
App Service, AKS, Azure SQL and Blob Storage for estates already standardised on Microsoft identity and tooling.
- Entra ID single sign-on against your existing directory
- Private endpoints so the database is never public
- Cost review at ninety days, once real usage is known
AWS
EC2, ECS, RDS and S3 where breadth of managed services or a specific region is the deciding factor.
- Infrastructure defined as code, not console clicks
- Multi-AZ database with automated failover
- Lifecycle rules that stop archive storage growing forever
SQL Server
The workhorse behind most of our ERP deployments — on-premise or Azure SQL, tuned for reporting alongside transactions.
- Index and execution-plan tuning on your real query load
- Always On availability groups where downtime has a price
- Point-in-time restore tested on a schedule
PostgreSQL
Our default for new multi-tenant platforms — row-level security, strong JSON support and no licence per core.
- Row-level security for fail-closed tenant isolation
- Partitioning for tables that grow past comfortable limits
- Logical replication for zero-downtime migrations
MongoDB
Document storage for the genuinely schema-variable parts — device payloads, audit captures and integration envelopes.
- Schema validation so flexible does not become formless
- Time-series collections for high-rate telemetry
- Used alongside a relational ledger, never instead of one
Data & Analytics
Reporting that reconciles with the ledger — one definition of margin, shared by finance, production and the board pack.
- Warehouse modelling with agreed, documented measures
- Scheduled distribution of reports to the people who act
- Drill-through from a dashboard tile to source transactions
Three Rules We Do Not Bend
They decide what goes into a build far more than any technology preference does.
Proven before promoted
A model or protocol reaches your production system after it has held up on your data, not because it is current.
One platform, many surfaces
Web, mobile, edge and reporting read the same ledger, so there is no second version of the truth to reconcile.
Degrades, never dies
Networks fail on plant floors. Capture buffers locally and reconciles on reconnect rather than dropping the reading.
What We Build With Day to Day
Not a logo wall — this is the set our engineers work in and support in production.
Want to Go Deeper
Than a Feature List?
Talk to the engineers who would build it — not a sales team reading a datasheet.