TensorIoT Start a pilot
Customer work · Selected engagements · Names withheld SECTION 03 / EVIDENCE

Systems that stayed running

Six tracks of production work, drawn from more than a hundred delivered engagements. Every entry replaced a manual process the organisation had stopped questioning. Customers are described by what they do, not by who they are.

Selected work — three at length 01

Opportunity, solution, outcome

Three engagements set out in full, to show what the shorter entries below are compressing.

Food processingComputer visionEdge ML

Spoilage caught before it ships

A large-scale food processor running high-volume production lines.

Opportunity

Quality control was manual. Identifying spoilage, tracking SKUs, and assuring shipment quality across high-volume lines was slow and error-prone, with millions of dollars of product loss at risk every year.

Solution

Machine learning deployed at the edge, tracking produce across production lines in real time. Vision models trained to detect spoilage indicators and flag non-conforming product before it reached shipment, with SKU-level insight surfaced to operations through a purpose-built dashboard.

Outcome

Approximately $50M in savings — product loss reduced, SKU tracking accuracy improved, spoilage caught before distribution. Quality control moved from manual bottleneck to automated, scalable process.

Consumer productsIoT platformPerformance

A device platform proven to a million

A global manufacturer of connected home safety and comfort products.

Opportunity

A new product line was about to launch on a platform that had never been tested at the scale the launch assumed. The existing architecture hit bottlenecks while simulating only ten thousand devices — against a roadmap calling for hundreds of thousands at launch and millions within three years.

Solution

A configurable, containerised load-testing harness able to simulate new-generation MQTT devices, older WebSocket devices, and mobile application users together — varying ramp rates, device mixes, and payloads to locate the precise architectural limits, followed by staged remediation against agreed service-level targets.

Outcome

The platform went from failing at 10,000 simulated devices to confidently supporting 300,000 MQTT devices, 30,000 WebSocket devices, and 10,000 concurrent mobile users — with a tested path to millions.

Financial servicesAgentic AIERP

Exceptions that resolve themselves

A major financial services firm.

Opportunity

ERP exceptions escalated to senior finance staff by default, pulling the most experienced people in the department into low-value matching work every close cycle.

Solution

An intelligent agent that detects exceptions across systems, categorises them by root cause, and resolves the routine ones autonomously — routing genuinely ambiguous cases to a human with the evidence already attached.

Outcome

85% of exceptions resolved automatically, and month-end close accelerated by five days.

Track 01 — Connected products & IoT platforms 02

Turning products
into platforms

Manufacturers who sell hardware increasingly have to sell a service around it. These engagements built the connectivity, the multi-tenant platform, and the update machinery that turns a shipped device into a product line with a roadmap.

Life sciencesIoT
Provider of scientific research equipment

Connectivity for lab instruments

A successful laboratory device hub had no scalable way to export its data or serve it to users remotely. We instrumented the devices, brokered telemetry to the cloud, and built for intermittent connectivity.

OutcomeDevices that previously had no connectivity now manage their own data with little or no internet, syncing curated telemetry when a link is available.
EnergyMulti-tenant SaaS
OEM leasing field assets to oil & gas operators

A platform their customers log into

The client wanted to offer telemetry and device management to the operators leasing their equipment. We built a multi-tenant application with tiered access, so each customer sees only their own assets while the client keeps an aggregate view.

OutcomeMoved from a pilot hardware phase to a scalable hardware-plus-software SaaS offering with a minimal DevOps footprint.
ManufacturingMigration
Commercial 3D printing manufacturer

Off a proprietary IoT stack

Tens of thousands of fielded devices across 23 models and 250 telemetry types sat on a licensed third-party IoT platform with limited data access. We rebuilt on open cloud services behind a façade API, so the existing front end never noticed.

OutcomeLicensing fees eliminated, full data availability recovered, and predictive maintenance became possible for the first time.
Consumer productsOTA
World leader in kitchen appliances

Smart appliances with real update control

A new line of connected products needed usage metrics, device health, and over-the-air updates. We built the full platform, including whitelisted provisioning and OTA targeting by region, model, or arbitrary metadata grouping.

OutcomeAn initial 10,000 connected appliances shipped, with additional units and models following immediately.
TransportationFleet
Mobile video surveillance provider

Analogue recorders into a live fleet

Video sat on on-premise recorders across a vehicle fleet with no way to extract analytics. We connected the existing hardware to the cloud and built a new multi-tenant platform giving both the client and their customers live device data.

OutcomeReal-time telemetry and analytics replaced after-the-fact retrieval, on infrastructure that scales with the business.
Consumer productsFleet
Electric bike-share systems provider

Docking stations as gateways

Bikes and charging hubs generated data nobody could reach, leaving the operator reliant on occasional ad-hoc pulls. We turned the existing docks into gateways that collect from docked bikes and push over-the-air updates back out.

OutcomePreviously isolated data became continuous insight into rider behaviour and asset health, with a route to ship new features remotely.
Public sectorSmart cities
Smart city lighting & utility technology company

A backend other manufacturers build on

The client needed a scalable backend, wrapped in an SDK, so third-party device makers could connect their own smart-city hardware. We built it serverless and bridged the existing municipal protocols rather than replacing them.

OutcomeNew customers onboard at any scale without fixed overhead — letting cities of any size adopt at their own pace.
AgricultureMulti-tenant SaaS
Commercial grow-light manufacturer

Camera pods that ship as a service

The client had built a camera pod that detects plant stress before it is visible to the eye, but no platform to connect the fleet. We delivered the IoT backend plus a multi-tenant front end with user, device, and business-intelligence functions.

OutcomeA SaaS offering their own customers manage directly, plus fleet visibility that guides the next generation of the product.
ManufacturingDevice management
Manufacturer of engines and power generation products

Onboarding and updating an engine fleet

The client needed device onboarding, monitoring, and update capability across a large fleet of engines with no ceiling on device count. We built provisioning, remote command, and OTA into a single operator interface.

OutcomeOne platform covering the whole device fleet, with no scaling limits designed in.
Track 02 — Industrial operations & asset intelligence 03

Plants, wells, buildings,
and the data they never gave up

Most industrial sites are already instrumented. The work is rarely adding sensors — it is getting data out of equipment that predates the cloud, some of it inherited through acquisition, without interrupting production.

ManufacturingPredictive maintenance
Diversified metals service business, operations in a dozen countries

Downtime predicted, not discovered

Eighty years of operating history and no operational data leaving the floor. We gathered telemetry from worksite machines and environmental sensors, then trained anomaly models and wired them into a work-order system.

OutcomeTimely maintenance notifications and unprecedented facility visibility, on an architecture that extends to a growing number of sites.
Construction materialsMining
Major producer of aggregates-based construction materials

Every site on one pane of glass

Visibility was needed across primary operations at multiple sites, over a wide variety of industrial equipment — some inherited through acquisitions — without modifying infrastructure or disrupting operations.

OutcomePreviously uncaptured data — running time, crusher utilisation, load levels, conveyor speeds — now drives live decisions and cross-site performance comparison.
Life sciencesContinuous process
Life sciences manufacturer, continuous production line

Root cause in hours, not weeks

Process controls captured telemetry, but exporting it was impractical and analysing it worse. We brokered the existing telemetry to the cloud, added external sensors, and served it through a UI with anomaly detection and configurable alerting.

OutcomeNew insight into the variables driving quality, and a drastic reduction in root-cause analysis time during quality investigations.
Oil & gasAnomaly detection
Large privately held oil and natural gas producer

Anomaly detection across thousands of assets

A distributed asset monitoring solution needed modernising, with the client wanting maximum value from the replacement. We added detailed asset modelling and per-asset automatic anomaly detection that switches on once enough training data has flowed.

OutcomeAnomaly detection rolled out across thousands of assets, with a comprehensive operational view that directly improved field efficiency.
EnergyModernisation
Unconventional oil and gas company

Off the on-premise historian

An expensive legacy on-premise system held field asset performance data hostage. We ingested from field assets and sensors into a data lake with a proper site and asset hierarchy, available to dashboards and analytics alike.

OutcomeNear real-time remote monitoring across the organisation — more assets managed per technician, and less overall downtime.
AutomotiveSustainability
Global tier-one auto parts supplier

Energy visible by area of the floor

Sustainability targets and cost pressure pointed at the same number: energy per unit produced. We built an industrial data lake and visualised utilisation across the factory floor to locate where the savings actually were.

OutcomeAn immediate 5% improvement in energy efficiency identified, with a route to 15% through further process integration.
Green energyOEE
Solid oxide fuel cell power generation

OEE across distributed generators

Generators deployed near the communities they power reported into an ageing historian. We replaced it with cloud connectivity from the edge gateway up, and built the business intelligence and alerting layer on top.

OutcomeAggregated visibility and near real-time OEE comparison across sites, plus a data architecture their scientists used to start building predictive models.
IndustrialAccelerator
Industrial asset modelling

Asset hierarchies built automatically

Modelling industrial assets by hand is tedious, slow, and error-prone. We built a converter that runs at the OT network edge and derives asset model hierarchies automatically from tags already present in existing systems.

OutcomeUp to 90% less effort configuring asset hierarchies and data connectivity — shortening time-to-value for OEE, defect detection, and ESG reporting alike.
Real estateSustainability
Large logistics real estate investment trust

Mesh sensor networks, finally readable

A global warehouse estate ran building monitoring over mesh networks that never reached the cloud. We built the edge decoding on the gateway and a high-volume delivery stream into a data lake.

OutcomeHeterogeneous sensor networks across a global property portfolio now report real-time operational conditions.
Track 03 — Computer vision 04

Cameras as
measuring instruments

Almost every deployment below ran on cameras the customer already owned. The value was never the hardware — it was finally being able to count, inspect, and verify what the hardware could always see.

AutomotiveDefect detection
Leading automotive parts manufacturer

Below the threshold of human vision

Certain forming defects on part assembly lines were being missed by visual inspection. We deployed vision systems at the critical inspection junctures, each with full edge resources for local inference and line integration.

OutcomeDetection accurate to roughly 1 mm — better than the human eye — running inside line cycle time and alarming instantly on a defect.
Food processingOffline-first
One of the world’s largest meat processors

Counting livestock without a network

Counting at transfer points is critical for inventory, compliance, and planning — and was done by hand at facilities with limited or no connectivity. We ran inference locally, no connection required, streaming results to a mobile app used on the floor.

OutcomeHigh counting accuracy across difficult lighting and crowding conditions, and substantial labour released from manual counting.
Food processingCompliance
Large-scale meat processing facilities

Sanitation, actually verified

Every worker had to perform each sanitisation step in the right order for adequate time, on every shift, with no practical way to confirm it. We trained a model to read body position, movement sequence, and time-on-task against each step.

OutcomeMeasurably improved compliance, plus identification of the specific shifts needing training — targeted intervention instead of blanket retraining.
Life sciencesVoice + vision
Global pharmaceutical manufacturer

Gowning guidance that watches back

Clinical workspaces depend on correct gowning procedure. We built an edge system of cameras, microphones, and a display that guides each entrant by voice while verifying compliance visually — and keeps working while disconnected.

OutcomeHands-free compliance assistance with automated, trackable metrics for internal reporting.
RetailBehaviour analytics
National fuel retail and convenience network

Forecourt and aisle, measured

Hundreds of sites, high throughput, and merchandising decided on intuition. Using the existing security cameras, we tracked vehicle arrival and dwell by zone, and in-store entry, occupancy, and traffic heat maps.

OutcomeTraffic correlated directly with campaigns, staffing and maintenance scheduled proactively, and layout decisions grounded in observed behaviour.
TelecommunicationsEdge / MEC
Global telecommunications operator

Driver safety at the network edge

Commercial fleets needed real-time detection of distraction, fatigue, and seatbelt non-compliance. Cloud round-trips were too slow, so inference runs on a local cellular edge node with alerts going straight back to driver and manager.

OutcomeNear real-time safety visibility fleet-wide, holding up in areas with limited connectivity.
ManufacturingCounting
Manufacturer of pallets and shipping containers

Counting goods as they are palletised

Manual counting at the wrapping station created bottlenecks and inventory discrepancies that rippled downstream. We gave the station 360-degree camera coverage feeding an edge device, counting continuously inside the existing workflow.

OutcomeManual counting eliminated, error rates down, inventory accuracy up, and labour released to other work.
Life sciencesQuality control
Manufacturer of spinal surgical implants

A quality gate for surgical parts

Inspection was fully manual in a zero-tolerance environment. We designed a controlled imaging compartment for consistent capture, plus an ingestion pipeline tagging every image with part numbers and ground-truth defect labels.

OutcomeThe data infrastructure and pipeline needed to screen automatically for chips, scratches, and anodisation faults — a bottleneck becoming an automated gate.
RoboticsGeospatial
Autonomous mobile robot manufacturer

Cleaning robots as environmental sensors

The robots cleaned well but could not recognise or map points of interest, limiting their value as a data platform. We integrated edge vision that classifies the environment during normal operation and syncs during charging intervals.

OutcomeSingle-purpose machines became a source of geospatial intelligence — better routing, and a new commercial offering.
Consumer goodsPlanogram
Leading global beverage company

Coolers that audit themselves

Display compliance was validated by people reviewing photographs taken during merchandiser visits. We specified the hardware and built an edge-to-cloud platform that collects cooler images automatically and runs planogram and out-of-stock detection.

OutcomeReal-time visibility into planogram compliance and stock position, supporting just-in-time delivery.
Food & logisticsLIDAR
Food and pet food conglomerate

Watching the truck get loaded

Management had no granular view of shipment loading or load balancing. We put edge hardware in the loading docks and on the forklifts, fusing camera and LIDAR data to track loading against the pick list as it happens.

OutcomeLoading verified while it happens rather than after, with live visibility into outgoing shipment progress.
Supply chainSustainability
Supply chain technology and packaging manufacturer

Scoring a distribution centre on materials

Their customers had public sustainability commitments and no measurement. We built vision on the outbound pallets to identify pallet type and product SKUs, producing a comparable sustainability score per site.

OutcomeDaily material-usage reporting with active suggestions for reducing carbon footprint and increasing reusable material.
Track 04 — Forecasting & decision models 05

Decisions that used to
be made on averages

Demand, prices, absence, traffic, routes. Each of these was already being forecast — in a spreadsheet, by someone experienced, using last year as the model.

Food processingPrice forecasting
One of the world’s largest meat processors

Livestock prices and production levels

Production strategy and price optimisation depended on forecasting a volatile market. We combined historical time series with external market, production, and sales data into a forecasting pipeline.

OutcomeForward visibility on sales and prices that lets the business calibrate against over- and under-production.
RetailDemand
Footwear, apparel, and accessories brand

Foot traffic, a month ahead

Inventory and workforce planning needed a view of store traffic that the existing process could not give. We built a data lake combining historical sales with holidays, events, promotions, and weather, then forecast from it.

OutcomeAccurate monthly traffic predictions across more than 60 stores, improving both inventory and workforce management.
ManufacturingDemand
Provider of scientific research equipment

Production scheduled to real demand

Manufacturing schedules were set without a reliable demand signal. We fed site-specific manufacturing data alongside historical sales, price, and promotional metrics into a demand forecasting model per site.

OutcomeMarkedly better elasticity to demand — less over- and under-production, optimised stock and inventory space, and reduced waste.
AutomotiveWorkforce
Multinational automotive manufacturer

Absence and attrition, forecast

Staffing manufacturing lines was persistently difficult, and the historic records were non-numeric. We transformed them into usable features, then forecast unplanned absence and attrition into a single dashboard.

OutcomeA staffing pipeline built on prediction rather than reaction — and an organisation that saw what else ML could address.
Supply chainRouting
Major foodservice supply chain

Drivers and routes, matched properly

Delivery scheduling and route planning were built on experience rather than analysis. We clustered drivers by performance for better allocation, and modelled the factors that actually drive route cost.

OutcomeExisting routes optimised and prospective route changes evaluated as a data-driven decision rather than a judgement call.
EntertainmentMLOps
Global television and sports entertainment company

Ad placement matched to programming

Advertising spots were matched to programming by hand. We built the recommendation models with the client’s subject-matter experts, plus an MLOps and retraining pipeline to handle drift over time.

OutcomeScaled across multiple brands, improving ad targeting, engagement, and conversion together.
Consumer productsPersonalisation
Leading skincare provider

Recommendations past the cold start

An existing recommendation flow ran on business rules and converted poorly for new customers. We ingested user input, historical, and clickstream data into a proper recommendation engine, exposed by API to the mobile app.

OutcomeBetter recommendations for cold-start and existing customers alike, and measurably better conversion.
AutomotiveEV fleet
Connected mobility solutions provider

Electric fleets have different physics

EV fleets brought new operational questions — daily state of charge, long-term state of health, charging cost, emissions. We built high-performance ingestion with EV-specific metrics and anomaly tracking.

OutcomeDrivers and service partners alerted on tyre pressure and battery anomalies, with per-vehicle efficiency compared across the fleet.
TechnologySelf-service ML
Platform-as-a-service company

Machine learning for non-specialists

The client wanted their own customers to get ML insight from data already in the platform, without hiring data scientists. We built an application exposing five model types behind a few well-chosen parameters.

OutcomeCustomers with no ML background build their own models and get daily, weekly, or monthly insight from their data.
Track 05 — Document AI & conversational systems 06

The work that arrives
as forms and phone calls

Claims, letters of credit, medical records, disputes, service requests. Work that used to be absorbed by whoever had the least standing to refuse it.

Financial servicesDocument AI
Legal-data platform serving investors

A 700,000-document backlog

Complex, inconsistent forms were being read and keyed by people before any of the data became useful. We replaced the pipeline end to end with automated extraction, parsing, and formatting into their database.

OutcomeHundreds of thousands of forms processed through a scalable intake engine, starting with the entire backlog.
InsuranceDocument AI
National insurance carrier

Claims forms read automatically

Thousands of insurance claim forms were processed by manual data extraction. We built automatic extraction triggered directly by their existing document workflow, writing straight into their database.

OutcomeFaster and more accurate claim processing — and a foundation the client now wants to extend into automatic fraud detection.
HealthcareRevenue cycle
Orthopaedics and prosthetics company

Identifying who actually pays

Products sold direct to patients were mostly paid by insurers, but messy human input made determining the payer hard — leaving a substantial number of transactions unpaid. We built compliant OCR and ML workflows over new and existing records.

OutcomeA substantial reduction in unpaid claims, plus structured data supporting prediction of whether approval criteria are met.
EnergyTrade finance
Major international coal exporter

Letters of credit into smart contracts

Trade transactions ran on letters of credit requiring manual processing and extraction before anything could move. We automated the extraction and used it to trigger execution on their transaction platform.

OutcomeA modernised, accelerated transaction platform with far less manual handling between document and settlement.
Life sciencesBrand compliance
World leader in serving science

Brand review without the back-and-forth

Branding approval was slow and the feedback subjective. We combined object detection, OCR, colour detection, and business rules into a pipeline that ingests collateral and returns each violation and its type via API.

OutcomeThe most common and most critical branding errors caught automatically, with the pipeline extensible to further brands.
Financial servicesContact centre
Leading provider of financial services

What customers are actually calling about

The client wanted better insight into customer contacts to guide product development. We transcribed recorded calls, ran entity and sentiment extraction, and surfaced issues and trends with deep-dive search.

OutcomeClear visibility into customer touchpoints, driving both product improvement and self-service tools that reduce call volume.
Financial softwareIVR
Leading financial software company

Natural language instead of a menu tree

Touch-tone routing extended handling time and misrouted enough callers to matter. We replaced parts of the platform with a natural-language IVR that lets callers state intent and route themselves.

OutcomeShorter total call time and fewer misrouted transfers, with a measurable lift in customer interaction quality.
Business servicesSelf-service
Business services company

Disputes filed without an agent

Common dispute types required manual collection of customer data and heavy agent involvement. We built a conversational flow handling the four most common credit dispute categories end to end over the phone.

OutcomeCustomers file disputes without an agent, and human agents concentrate on the genuinely complex cases.
Public sectorCivic services
Conversational interface for citizens

City services over SMS

Citizens needed a single around-the-clock contact point to get information and report problems. We built an SMS platform with a deep conversation map that responds in natural language and opens service tickets automatically.

OutcomeEngagement in the client’s customer cities almost doubled, with more citizens reporting problems and receiving information.
HealthcareScheduling
Telemedicine service provider

Appointments booked by conversation

Scheduling in-home healthcare depended on outbound calling. We combined mass SMS with a chatbot that handles dialogue and takes real actions — adding to call lists, sending and changing appointment details.

OutcomeHigher member engagement and appointment scheduling, plus better-qualified follow-up from captured intent.
Life sciencesVoice
Provider of scientific research equipment

A lab assistant you talk to

Sterile laboratories are touchless environments — gloves on, no keyboards, everything on paper afterwards. We integrated voice services with the lab devices and built the interactions the work actually needs.

OutcomeLab staff command devices and receive readouts, compositions, and next steps without removing their gloves.
HealthcareTriage
Neurological disease research organisation

Self-assessment at population scale

A new method for detecting early signs of a neurological condition created demand the organisation had no way to serve. We built the assessment chatbot, the supporting site, and the contact centre behind it.

OutcomeRemote self-assessment with onward resources, plus analytics on drop-off and technical issues to improve future engagement.
Track 06 — Data, migration & platform engineering 07

The work underneath
the work

Reporting, migration, load testing, disaster recovery, support. None of it makes a good demonstration. All of it decides whether the rest survives its first year.

SemiconductorReporting
Leading semiconductor manufacturer

Reports in minutes, not weeks

Financial reporting ran on a slow desktop-based process. We built an end-to-end automated data-lake reporting solution, delivered as a deployable package, and trained their team on it as we went.

OutcomeReport generation cut from weeks to minutes, plus a repeatable template for building new reports at the same speed.
EntertainmentData lake
Lottery solutions provider

Data arriving faster than it could be cleaned

Data production had outpaced largely manual aggregation and cleanup. We designed and deployed automated ingestion that integrates uploads into a data lake and unlocks visualisation directly.

OutcomeIngest, transform, store, and visualise at scale — in less time and for less money than the manual process it replaced.
SustainabilityTime series
Global technical products and services provider

Security review before production

An innovation-stage solution relying on enormous sensor volumes was about to go to production. We ran a well-architected review, closed the security gaps in code and integration points, and rebuilt the calculation platform on a proper time-series store.

OutcomeReduced time and cost of data processing, on a materially stronger security posture.
AgricultureConsolidation
Global manufacturer of irrigation products

Two environments into one

IoT-enabled irrigation controllers ran on a hybrid on-premise and cloud setup across two environments. We consolidated end device, cloud infrastructure, and front end into a single environment serving both commercial and contractor product lines.

OutcomeLower infrastructure cost, dramatically improved app responsiveness, and a foundation that supports the next generation of products.
Consumer productsDevOps
Smart video doorbell manufacturer

Infrastructure someone watches at 3am

The client wanted a scalable DevOps pipeline for managing and monitoring their stack, and direct support for what was already deployed. We documented and optimised the existing code, then automated the deployment path.

OutcomeA modernised, automated backend backed by 24/7 second- and third-line monitoring and support.
LogisticsEmergency response
Engineering, medical, and emergency management consultancy

Inventory that changes by the hour

Disaster response makes supply chain logistics unpredictable: constantly evolving inventory, minimal setup time, and critical metadata like vaccine lots and expiry dates. We connected warehouse workers’ phones to a scanning system integrated with purchase orders.

OutcomeReal-time visibility of a disparate and shifting inventory, reducing shrinkage and misplacement.
AutomotiveTrack and trace
Leading automotive manufacturer

Lean manufacturing needs a live ETA

Disruption from incoming parts shipments had been identified as an addressable cause of loss. We integrated ERP parts data, truck loading and departure events, and GPS telemetry into a single view.

OutcomeLive ETAs across the inbound supply chain, allowing operational adjustment in advance, with geofencing improving logistical accountability.
Food processingTraining
One of the world’s largest meat processors

Onboarding in an industry with high turnover

Training and onboarding had to be repeatable and engaging in a sector where turnover is constant. We built an immersive 3D training application covering manual tasks including machine maintenance and correct equipment use.

OutcomeLower training cost, less turnover, and more consistency across production processes.
Consumer productsRemote operations
Environmental field operations team

Audits without the flight

Regular audits required in-person visits to fulfilment centres. We built a responsive application with live video and integrated assessment forms, so auditors and site managers work through the audit together, remotely, within IT security constraints.

OutcomeAudits conducted remotely from one application — higher productivity, lower cost, and less travel.

Watch Customer names are withheld throughout. Where an engagement is publicly referenceable we are glad to talk about it directly — ask, and we will tell you which of these we can name.

Next step 08

Your version of this,
in eight weeks

One use case, built on your data, against a KPI agreed before the first commit.

Start a pilot → How delivery works →