Case studies

What we built, and what it changed.

Every story here starts with a problem a real team had, on a plant floor, in a product, or in the systems around them. It says what we built, and ends with what is different now. Start with the one that sounds like yours.

Start with the featured story ↓ Browse them all
All case studies

Pick the one that sounds like yours.

Each card carries the client, the problem, what we built, what changed, the facts we can stand behind, the stack, and whether it is still running.

Manufacturing & engineeringFortune 50
The demo everyone remembers is the garage door opening before the car turns in. What sold it internally was that the override always wins.
Product owner, connected services, automotive OEM
Connected carSmart homeGeofencingIoTCloud
Automotive OEM · Connected car · IoT

A car that sets the house before it arrives

For a Fortune 50 automotive OEM: a connected-car-to-smart-home bridge. The vehicle’s position against a home boundary drives presets for lights, HVAC and the garage door, through a home control unit and a cloud layer, either automatically or once the owner confirms.

  1. 01The problem The car knew where it was. The house did not, and the owner was setting lights, climate and the garage door by hand on the way in.
  2. 02What we built Three layers: the app, a home control unit and a cloud service. Presence resolved between two phones per home, and a manual override that always wins.
  3. 03What changed Lights, climate and the garage door set themselves as the car crosses the home boundary, or wait for a tap.
Fortune 50automotive OEM3 layersapp, control unit, cloud2 phonesper home, presence resolved
Shipped · app, control unit, cloud Read the case study →
Manufacturing & engineering
For the first time the limit on the screen is the limit the equipment can actually take today.
Head of inspection engineering, European engineering enterprise
Sensor dataEquipment wearSafe operating limitsProcess plants
Engineering & inspection · Enterprise · Europe

Safe operating limits for chemical and refinery plants

For a European engineering and inspection enterprise: sensor thresholds tracked against equipment wear, so the limit a plant runs to reflects the state the equipment is actually in.

  1. 01The problem A safe limit set at commissioning stops being safe as the equipment wears, and nobody was recalculating it.
  2. 02What we built Sensor readings tracked against the wear state of each piece of equipment, with the operating limit recalculated to follow it.
  3. 03What changed The limit a plant runs to now reflects the equipment as it is, not as it was installed.
7+ yrsworking relationship, and countingSensor → limittracked against wearChemical · refineryplants
In production Read the case study →
Manufacturing & engineering
We stopped arguing about which plant’s number was right. There is one now, and the drawings are searchable.
Group master data lead, automotive components group
Master data managementEngineering drawingsDigitizationSearch
Automotive components · Global group

One master data layer, and drawings that are no longer retyped

For a global automotive components group: master data management across plants and product lines, and the engineering drawings that used to be keyed in by hand now digitized, structured and searchable.

  1. 01The problem Every plant and product line kept its own master data, and engineering drawings were retyped by hand.
  2. 02What we built One governed master data layer across plants and product lines, and a pipeline that digitizes drawings into structured, searchable records.
  3. 03What changed One version of the truth across the group, and drawings anyone can find.
5+ yrsworking relationship, and countingPlants × linesone master data layerDrawingsdigitized, searchable
In production Read the case study →
Manufacturing & engineering
Dealers raise a request against the serial number and can see where it is. The phone stopped ringing about status.
Service director, industrial automation maker
Installed baseService requestsSpare partsOTPAdmin panel
Industrial automation & robotics · Service platform

Service requests for CNC, robots and lasers, from the floor

For a global industrial automation maker’s customers: register machines by model and serial, raise service, spare-part and repair requests against the installed base, confirm out-of-warranty charges by OTP, and track every request through to closure.

  1. 01The problem Service, spare-part and repair requests came in by phone and email, with no record against the machine they were about.
  2. 02What we built A service platform for end users, makers and dealers: machines registered by model and serial, requests raised against them, OTP confirmation for out-of-warranty charges, an admin panel for the back office.
  3. 03What changed Every request is tracked from the floor to closure.
3 audiencesend users, makers, dealersOTPout-of-warranty charges confirmedTo closureevery request tracked
In production · end users, makers, dealers Read the case study →
Manufacturing & engineering
We went into the board with one number and a model that had already been right once. That is a different conversation.
Head of engineering computing, global engineering group
SnowflakeSnowparkPythonPower BIDiscrete-event simulation
Engineering group · HPC · Global

How many cores do we actually need?

A global engineering group runs crash, structural and fluid simulations overnight. Utilisation looked low at 54%, and the queue still hit ten-hour waits. We replayed the job logs through a simulation so the next hardware decision came down to one number.

  1. 01The problem Average utilisation hides queueing near capacity. Leadership needed a defensible sizing method, not a hunch.
  2. 02What we built A medallion pipeline in Snowflake feeding a discrete-event simulation on Snowpark, with a Power BI dashboard and a companion web app for scenario runs.
  3. 03What changed The model predicted the 2.56-hour wait the real cluster then delivered, and found the point past which more cores stop helping.
10.25 h → 2.56 h95th-percentile wait, before and after2.56 hpredicted, then observed, on unseen data~1,500cores: the saturation point
Delivered · re-validated quarterly Read the case study →
Manufacturing & engineering
The Monday report used to take a person a day. Now it is there at six in the morning, and it is the same for everyone.
BI lead, European automotive supplier
SSISPower BIETLData
50,000-person automotive supplier · Europe

ETL and Power BI reporting for a large automotive supplier

Manual Excel reporting replaced by a structured SSIS pipeline and Power BI dashboards, with role-based access and data refreshed every six hours.

  1. 01The problem Reports built by hand, in spreadsheets, for a 50,000-person organisation.
  2. 02What we built SSIS ETL into a reporting model, Power BI dashboards, role-based access.
  3. 03What changed A six-hour refresh and reporting that scales with the business.
50,000people in the organisation6-hourdata refreshRole-baseddashboard access
In production Read the case study →
Financial services
It is rare to see an external team develop such a strong understanding of both the business requirements and the underlying technical architecture in such a short period of time.
CEO, B2B wealth-planning platform
LangGraphClaude Sonnet 4.5Bedrock AgentCoreFastAPIPostgreSQL 16
WealthTech · B2B SaaS · North America

From an AI proof-of-concept to a bank-deployable assistant

A wealth-planning platform had a demo its users loved and nothing under it a bank would sign off: no permission model, no audit trail, answers that changed between runs. We rebuilt it as a grounded, explainable, multi-agent assistant.

  1. 01The problem The PoC proved demand and could not survive contact with a regulated buyer.
  2. 02What we built Three services: an orchestrator with five domain specialists on LangGraph, an MCP tool server exposing 12 tools over the planning engine, and a history and presence service. Moved from EKS to Bedrock AgentCore.
  3. 03What changed Five planning scenarios live, every answer grounded in the real planning engine, RBAC on every call, and a confirmed-write flow for changes.
3independently deployable services1 + 5orchestrator and specialists12MCP tools over the platform APIs
In production · milestone 1 Read the case study →
Insurance
LangGraphQdrantClaude Sonnet 4.5FastAPIReact 18
InsurTech · Home risk data

From a stack of home documents to a source-traced property record

Inspection reports, contractor invoices, manuals and warranties, none of them in the same shape. A LangGraph pipeline turns them into structured, linked records, with every field traced back to the page it came from.

  1. 01The problem Four entity types tangled across documents, duplicates everywhere, and a manual review that did not scale.
  2. 02What we built An eight-node LangGraph state machine: classify chunks, extract per entity type, merge, validate against schema, assign sources, resolve relationships.
  3. 03What changed One traced, unattended run over a whole document set, where a reader used to spend an hour or more per set.
8orchestrated nodes4extraction agents, one per entity typeEvery fieldtraced to document, page, section
PoC delivered · extends without re-architecting Read the case study →
Media & communications
LangGraphSQSAppSyncQdrantNext.jsFastAPI
AI platform · Communications teams

From a single chat box to twenty-five services doing the work

Communications teams wanted monitoring, risk tracking, research and content behind one chat surface, and none of it blocking the rest. Everything runs asynchronously: requests cross a queue, results come back over a socket.

  1. 01The problem On-demand chat and on-schedule monitoring pull in opposite directions, and one slow job used to hold the whole conversation.
  2. 02What we built 25 independent services across eight layers: four conversational AI services, discovery and analysis services, exploration agents, knowledge providers over Qdrant and the live web.
  3. 03What changed New assistants ship as validated YAML, long research jobs no longer hold a request open, and a failure stays local to one capability.
25independent services17stages in the deep-research pipeline5assistant types from configuration alone
Live · monitoring migration continuing Read the case study →
Security
Claude Agent SDKMCPAWS ECSLangfuseGitLab
Container security · 40+ language ecosystems

Automated CVE research for a container-security platform

Every disclosure raised the same three questions: does a fix exist, does it apply to our versions, can it be backported. Seven agents now sweep NVD, MITRE, repos, advisories and exploit databases and hand back a verdict with its evidence.

  1. 01The problem CVE research was fragmented across sources and gated on the few engineers who could do it defensibly.
  2. 02What we built Seven agents in a fixed sequence on the Claude Agent SDK, 16 MCP research tools, event-driven on ECS, with a reviewer UI and Langfuse tracing.
  3. 03What changed Research runs continuously, expert time moves to judgement, and the verdicts feed the patch, test and ship agents downstream.
7agents in sequence16MCP research tools40+language ecosystems covered
Delivered · feeds the patching agents Read the case study →
Security
Amazon EKSMCPGitLabJiraLangfuse
DevSecOps platform · Remediation

An agent that backports security fixes on its own

Knowing the fix was never the slow part. Turning it into a patch that applies to a customer codebase that has drifted for years was, and it waited on scarce senior engineers. An eight-phase agent does it now, in parallel, across many codebases.

  1. 01The problem Backports were manual, one at a time, and coverage was judged by eye.
  2. 02What we built Eight phases from project analysis and AST hunk extraction to dependency-aware application and self-validation, on EKS, with an MCP server over GitLab and Jira.
  3. 03What changed Runs without a human in the loop, early-exits on already-patched code, and flags or fills the patterns it missed.
8autonomous pipeline phases3outputs per run: verdict, changes, patchParallelacross many codebases at once
In production · wider language coverage next Read the case study →
Security
Agentic AILLMMulti-agentAutomation
Security teams · Vulnerability triage

An autonomous multi-agent system for CVE research

Analysts were triaging CVEs by hand, one source at a time. A multi-agent research system now validates each vulnerability across sources and turns hours of triage into minutes, thousands of CVEs at once.

  1. 01The problem End-to-end triage took hours per CVE and did not scale with the rate of disclosures.
  2. 02What we built Specialised agents for discovery, cross-source validation and reporting, orchestrated so thousands of CVEs run at the same time.
  3. 03What changed Triage that took hours completes in minutes, with the evidence attached to every finding.
Hours → minutesper CVEThousandsof CVEs processed at onceMulti-sourcevalidation on every finding
Delivered Read the case study →
Financial services
OCRLLMOrchestrationAutomation
Finance teams · Accounts payable

OCR orchestration that matches invoices to purchase orders

Adaptive OCR reads the invoice, a language model checks it against the purchase order, and the finance team sees only the exceptions.

  1. 01The problem Invoice verification was manual, error-prone and slow, and every vendor’s layout was different.
  2. 02What we built An orchestration layer that picks the OCR path per document, extracts line items and reconciles them with a language model.
  3. 03What changed Fewer manual errors, a lower cost per invoice, and a faster close.
AdaptiveOCR per document layoutLLMline-item reconciliationExceptions onlyreach the finance team
Delivered Read the case study →
Enterprise
LLMRAGCopilotEnterprise
Large enterprises · Knowledge transfer

LLM-powered assistants that capture what the experts know

Tribal knowledge lives in people who leave. Assistants trained on the organisation’s own material make it searchable, answerable and part of onboarding.

  1. 01The problem Expertise locked in individuals and scattered across documents nobody could find.
  2. 02What we built Assistants over the organisation’s corpus with retrieval, citations and access control by role.
  3. 03What changed Faster onboarding, and expertise that stays when people go.
Onboardingmeasurably fasterCitedanswers from the corpusRole-basedaccess to what each person may see
Delivered Read the case study →
Enterprise
CopilotKnowledgeSearch
Teams at scale · Knowledge sharing

Simplifying how a team shares what it knows

AI knowledge tools that help teams onboard faster and keep institutional expertise reachable, whatever the size of the organisation.

  1. 01The problem Answers existed, somewhere: in someone’s head, or an old thread.
  2. 02What we built A knowledge layer over wikis, tickets and chat with an assistant on top of it.
  3. 03What changed Critical information reachable by the people who need it, when they need it.
Wikis · tickets · chatone knowledge layerAny scaleof organisationAssistanton top of all of it
Delivered Read the case study →
Consumer products
NLPVoice AIAndroidEdge
Global electronics brand · Children’s tablets

A voice assistant for a children’s tablet that works offline

A keyword wakes the assistant, a curated library answers offline, and search takes over online. Built for a global electronics brand’s children’s Android tablets, shipped across 70+ countries.

  1. 01The problem Children ask questions a general assistant should not answer, and the tablet is often offline.
  2. 02What we built An on-device keyword trigger, NLP over a curated Q&A library, and an online search fallback with guardrails.
  3. 03What changed Voice interaction that works without a connection and stays inside a library the brand controls.
70+countries shippedOfflineQ&A from a curated libraryKeywordtriggered, on the device
Shipped Read the case study →
Mobility & IoT
Every incident now comes with a clip, a position and a verified driver. That changed how we talk to insurers.
CTO, fleet technology operator
Edge inferenceKafkaNestJSKinesis VideomTLSMongoDB
Fleet technology · Long-haul, last-mile, passenger

A fleet platform that verifies the driver, not just the vehicle

Telematics said where the vehicle was and nothing about who was driving, how, or what happened. We built the edge-to-cloud stack: a biometric check at ignition, ADAS and driver-monitoring inference on the device, and collision footage kept without a live video link.

  1. 01The problem No driver identity, no behaviour signal, and no evidence when something went wrong.
  2. 02What we built On-device inference and evidence capture with a three-tier fallback store; mTLS ingress, Kafka, five NestJS services, MongoDB, Redis and S3 behind it; video on demand over WebRTC.
  3. 03What changed Every incident now produces a clip, a position, a timestamp and a verified driver, with no telemetry or video lost under bad 4G.
5core microservices3storage tiers per clip0%telemetry or video lost on erratic 4G/5G
Live with a commercial fleet Read the case study →
Insurance
GCPCloud ArmorAPI GatewayTerraformBigQueryCloud Functions
InsurTech analytics · Google Cloud

A public filing API with the database three layers back

Rate-filing data that partners wanted to pull by machine, with no network path from the endpoint to the database and no credential readable by anyone who does not need it. A serverless API on Google Cloud where the security lives in the topology.

  1. 01The problem Expose valuable filing data to partners without exposing the database or a single long-lived key.
  2. 02What we built A global load balancer with Cloud Armor, API Gateway validating short-lived JWTs, Python Cloud Functions in a private VPC, secrets fetched at call time. All of it in Terraform.
  3. 03What changed Partners get keys and responses with no manual step in between. The page reports no performance figure because none has been earned yet.
3defence layers before any logic runs0manual steps from key issue to responseTerraformfor every resource in the estate
Phase 1 delivered · Apigee next Read the case study →
Education
Cloudflare R2AWS CloudFrontMigrationArchitecture
EdTech NGO · Content delivery

Cloud delivery costs cut by 97% with a CloudFront to Cloudflare move

An education non-profit was paying CloudFront egress on every video its learners watched. We moved delivery to Cloudflare R2 with no downtime and no egress fees, and the bill fell by 97%.

  1. 01The problem A content bill that grew with every learner, most of it egress.
  2. 02What we built A staged migration of the media library to R2 behind Cloudflare, with the origin switched over without a pause in service.
  3. 03What changed Zero downtime during the move and zero egress fees after it.
-97%delivery cost0downtime during the moveZeroegress fees after it
Delivered Read the case study →
Education
LMSRole-based accessSchedulingEdTech
Robotics training provider · LMS

A three-portal e-learning platform for robotics training

Students, faculty and admins each got their own portal on one platform: role-based access, course and batch management, live class scheduling and progress tracking.

  1. 01The problem Robotics courses run in batches with live sessions, and off-the-shelf tools did not fit the way the school taught.
  2. 02What we built Three portals over one backend, with batches, schedules, live classes and progress reporting.
  3. 03What changed One system from enrolment through certification, run by the provider’s own admins.
3portals: student, faculty, adminBatchescourse and cohort managementLiveclass scheduling built in
Delivered Read the case study →
Mobility & IoT
They are on the queue when our customers are, and they know the platform. That was the whole brief.
Head of digital products, marine and energy group
SREIoTSupportInfrastructure
Marine, energy & certification group · Europe

Production support for a European IoT fleet platform

A 12-person dedicated team runs L1 to L3 support, deployment, configuration and infrastructure for a world-leading marine, energy and certification company’s IoT fleet products, on central European hours.

  1. 01The problem A fleet platform in production needs people who know it, awake when its customers are.
  2. 02What we built A dedicated team covering incident response, releases, configuration and infrastructure, with the runbooks to match.
  3. 03What changed One team and one queue, from first response to root cause.
12people on the dedicated teamL1 · L2 · L3all three lines of supportCETcentral European working hours
Ongoing support Read the case study →
Mobility & IoT
Azure FunctionsServerlessEvent-driven
Marine IoT fleet platform · Europe

From IaaS to serverless for a marine IoT platform

Sensor data arrives in bursts, and fixed servers were either idle or behind. We moved the platform to Azure Functions so it scales with events and costs what it uses.

  1. 01The problem Over-provisioned for the peaks, under-provisioned when the whole fleet reported at once.
  2. 02What we built Event-driven functions for ingest, processing and alerts, replacing the VM tier.
  3. 03What changed No more sizing guesses, a lower run cost, and scaling that follows the sensors.
Event-drivenscaling on sensor burstsVMs → Functionsthe compute tier replacedLoweroperating cost, no idle servers
Delivered Read the case study →
Financial services
WhatsAppNode.jsKubernetesMicroservices
Indian mutual fund · 6 million investors

Customer support on top of WhatsApp for six million investors

Investors already had WhatsApp open. We put account information and a route to a support agent there, on a Node.js microservices platform managed on Kubernetes.

  1. 01The problem Support channels investors had to learn, and a scale that broke monoliths.
  2. 02What we built Node.js microservices behind the WhatsApp Business API, on Kubernetes, with session state and agent handover.
  3. 03What changed Balances, statements and an agent, from the app people already use.
6Mactive investors servedKubernetesmanaged microservicesAgenthandover inside the chat
In production Read the case study →
Education
LaravelPHPModernisationEdTech
EdTech · E-learning & certifications · Europe

A legacy PHP platform re-engineered to Laravel, module by module

An e-learning and certifications platform on an old PHP Yii stack, migrated to Laravel one module at a time in two-week sprints, with production never interrupted.

  1. 01The problem A framework nobody wanted to build on any more, and a live platform that could not stop.
  2. 02What we built A module-by-module migration plan, each sprint shipping one piece to production beside the old one.
  3. 03What changed A modern stack, with zero disruption to the operations it replaced.
2-weeksprints, one module each0disruption to productionYii → Laravelthe framework replaced
Delivered Read the case study →
Education
Dedicated teamEdTechStartup
EdTech startup · Europe

A dedicated product team for a European EdTech startup

A startup that needed to ship faster than its hiring allowed. We stood up a dedicated offshore engineering team that works as its own.

  1. 01The problem Product ambitions bigger than the local hiring market.
  2. 02What we built A dedicated team on the startup’s own backlog, ceremonies and standards.
  3. 03What changed Development capacity that scales with the roadmap instead of the job market.
Dedicatedteam on the startup’s backlogOffshorewith direct managementSharedsprints and ceremonies
Ongoing Read the case study →
Education
We onboarded a quarter of a million learners without an outage. Six months earlier we could not hold two hundred.
Programme director, education non-profit
ScalingPerformanceCloudRe-architecture
QUEST Alliance · Quest App · Backed by Accenture

An EdTech platform scaled 50×, from 200 to 10,000+ concurrent users

The Quest App had to take 250,000 learners onboarding at once. We found the bottlenecks, re-architected for load and got there with no downtime.

  1. 01The problem A platform that fell over at 200 concurrent users, with a quarter of a million learners on the way.
  2. 02What we built Load profiling, re-architecture for horizontal scale and a staged rollout.
  3. 03What changed 10,000+ concurrent users and 250,000 onboarded, with zero downtime.
200 → 10,000+concurrent users250,000learners onboarded0downtime through the scale-up
Delivered Read the case study →
Education
MicroservicesLaravelAngulari18n
EdTech · Five-year-old platform

A five-year-old EdTech monolith moved to microservices

Frontend re-engineered in Angular, PHP Laravel APIs refactored into services, multilingual support added, and integrations opened at platform level.

  1. 01The problem Five years of a single codebase, hard to scale and harder to change.
  2. 02What we built Service boundaries drawn from the domains, an Angular frontend, refactored Laravel APIs, i18n throughout.
  3. 03What changed Independent services that scale and ship on their own, and a platform others can integrate with.
5 yearsof monolith decomposedAngularfrontend re-engineeredMultilingualsupport added
Delivered Read the case study →
Education
GamificationCloudMobileEdTech
QUEST Alliance · Youth employability

A cloud-based, multi-device gamified learning platform

Twenty-first-century employability skills for young people, delivered as a mobile-first game they can pick up anywhere, on any device.

  1. 01The problem Learners without desks or fixed hours, and content that had to hold their attention.
  2. 02What we built A cloud platform with a gamified curriculum, progress and rewards, on mobile first and every device after.
  3. 03What changed Learning that fits around a young person’s day rather than the other way round.
Multi-deviceone account, any screenGamifiedprogress and rewardsMobile-firstbuilt for anytime learning
In production Read the case study →
Financial services
AWSAuto-scalingRedisELK
FinTech lender · United States

A FinTech lending platform scaled for growth

A US lending platform that needed to hold 2,000+ concurrent users and stay up. Horizontal auto-scaling, load balancing, Redis caching and ELK monitoring got it there.

  1. 01The problem Growth the platform could not absorb, and outages the business could not.
  2. 02What we built AWS auto-scaling groups behind load balancers, Redis for hot data, ELK for application monitoring.
  3. 03What changed 2,000+ concurrent users and 100% availability since.
2,000+concurrent users100%availabilityRedis · ELKcaching and monitoring
In production Read the case study →
Financial services
WhatsAppMicroservicesScale
Axis Mutual Fund · 8.7 million investors

A WhatsApp support platform for 8.7 million investors

Account information and support agents inside WhatsApp for one of India’s largest mutual funds, on a microservices architecture built to take the load.

  1. 01The problem Millions of investors, one channel they all use, and a backend that had to keep up.
  2. 02What we built Microservices behind the WhatsApp Business API with routing to live agents.
  3. 03What changed Self-service at scale, and agents for the rest.
8.7Mactive investorsWhatsAppas the support channelMicroservicesbuilt for the load
In production Read the case study →
Financial services
LendingKYCIntegrationsFinTech
FinTech lender · Low-cost lending · United States

A loan origination and servicing platform for low-cost lending

KYC, credit scoring, disbursement, collections: a full platform for a US lender, integrated with 14 external systems, and a dedicated team still building it.

  1. 01The problem Low-cost lending only works when the whole flow is automated.
  2. 02What we built Origination and servicing end to end, with 14 integrations for scoring, verification and collections.
  3. 03What changed A lending operation that runs on the platform, with a team that keeps it moving.
14external systems integratedKYC → collectionsthe full lifecycleDedicatedteam for ongoing development
In production Read the case study →
Financial services
SRESupportAWSPHP
Low-cost lending platform · PHP on AWS

Production support for a loan origination platform

Application support, infrastructure management and L1 to L3 incident resolution for a lending platform on PHP and AWS, under an annual maintenance contract.

  1. 01The problem A lending platform cannot wait for the next sprint when something breaks.
  2. 02What we built A support desk with escalation to engineers who know the code, plus infrastructure ownership.
  3. 03What changed Incidents resolved at the right line, and infrastructure kept current.
L1 – L3incident resolutionAMCannual maintenance contractPHP · AWSthe estate supported
Ongoing support Read the case study →
Financial services
TerraformGCPMicroservicesDevOps
FinTech payroll-lending startup · Google Cloud

Infrastructure as code on Google Cloud for a FinTech startup

A multi-tenant SaaS on domain-driven, reactive microservices, with every piece of infrastructure in Terraform and a DevOps setup that does not care which cloud it is on.

  1. 01The problem A startup that needed enterprise-grade infrastructure without an infrastructure team.
  2. 02What we built Terraform for the whole estate, a multi-tenant architecture, multilevel security, cloud-independent pipelines.
  3. 03What changed Environments that rebuild from code, and a platform ready for its second cloud.
100%of infrastructure in codeMulti-tenantSaaS architectureCloud-independentDevOps
Delivered Read the case study →
Real estate & built environment
Digital twinSimulationReal-timeSmart city
Qi Square · BtrLyf · Built environment

A high-performance digital twin for the built environment

Real-time building simulations, energy benchmarking, crowdsourced smart-city data and a place for every stakeholder to work on the same model.

  1. 01The problem Building performance data scattered across owners, operators and cities.
  2. 02What we built The BtrLyf platform: a simulation engine, benchmarking, data crowdsourcing and collaboration.
  3. 03What changed A live twin that stakeholders across the ecosystem share.
Real-timebuilding simulationEnergybenchmarking built inCrowdsourcedsmart-city data
In production Read the case study →
Travel
MVPiOSAndroidML
Travel & tourism startup · iOS and Android

An AI-powered travel app, from zero to beta in six months

Native iOS and Android apps with ML-based personalisation and the animations a consumer product needs, built from scratch and in beta within six months.

  1. 01The problem A startup with an idea and no product.
  2. 02What we built Native apps, a recommendation model, and the backend behind both.
  3. 03What changed A beta in users’ hands in six months.
6 monthsto beta launchiOS + Androidnative appsMLpersonalised recommendations
Shipped · beta Read the case study →
Software & agencies
.NET CoreKubernetesDevOpsMigration
Communications workflow software

A .NET monolith moved to .NET Core microservices on Kubernetes

A legacy communications workflow platform split into services that scale on their own, with in-memory caching and a release cadence the business could feel.

  1. 01The problem A monolith where every change was a full release.
  2. 02What we built .NET Core services on Kubernetes, in-memory caching, independent deployment.
  3. 03What changed Services that scale independently and features that ship faster.
.NET → .NET Corethe runtime movedKubernetesorchestrated servicesIndependentscaling per service
Delivered Read the case study →
Agriculture
KeycloakIAMSecurityMulti-tenant
Stellapps · AgriTech · Multi-tenant

Identity and access management for a multi-tenant AgriTech platform

One login across every application, role-based access, and a central admin dashboard, on Keycloak.

  1. 01The problem Several applications, several logins, and no single view of who could do what.
  2. 02What we built Keycloak as the identity hub, roles mapped across apps, a central admin console.
  3. 03What changed Unified login and role control across the platform.
Keycloakidentity and access hub1 loginacross the applicationsRole-basedaccess, centrally managed
In production Read the case study →
Consumer products
IoTWearablesMicroservicesOffline-first
BRL Outdoors · Wearable device

A microservices and IoT architecture for a wearable device startup

Voice-activated data capture, GPS tracking, an app that works offline and syncs when it can, and the cloud behind it, designed for a fishing wearable.

  1. 01The problem A device used far from signal, and a product built on what it records.
  2. 02What we built Microservices for ingest and sync, offline-capable mobile apps, GPS and voice capture.
  3. 03What changed Fishing intelligence that syncs when the boat gets back.
Offline-capablemobile appsGPS + voicecapture on the deviceMicroservicescloud sync
Delivered Read the case study →
Real estate & built environment
Dedicated teamDevOpsPropTech
Cloud real-estate loyalty platform

A dedicated product team for a real-estate loyalty platform

A cross-functional team for ongoing product development, architecture upgrades, DevOps and production support, embedded with a cloud loyalty platform for real estate.

  1. 01The problem A product that never stops needing engineers.
  2. 02What we built A cross-functional team owning features, architecture and operations.
  3. 03What changed A partnership rather than a project.
Cross-functionalproduct, DevOps, supportOngoingarchitecture upgradesEmbeddedwith the client team
Ongoing Read the case study →
Software & agencies
Dedicated teamStaff augmentationDelivery
Technology service provider · Europe

A dedicated offshore team for a European technology service provider

Delivery capacity across multiple projects, directly managed by the client, in an agile model that scales up and down with the pipeline.

  1. 01The problem More client work than the local team could take.
  2. 02What we built A dedicated offshore engineering team under the provider’s own management.
  3. 03What changed Capacity that flexes with demand.
Multipleprojects in parallelDirectly managedby the clientAgilecollaboration model
Ongoing Read the case study →
Software & agencies
Extended teamAgencyDelivery
Award-winning product agency · New York

An extended engineering team for a New York product agency

Collaborative and independent development across several client projects, on time and to spec, for an award-winning agency.

  1. 01The problem An agency whose demand outran its bench.
  2. 02What we built An extended team that plugs into the agency’s projects, from joint sprints to standalone builds.
  3. 03What changed Projects delivered on time and to spec, repeatedly.
Extendedteam, agency-managedOn timeand to specMultipleclient projects
Ongoing Read the case study →
Non-profit
TalendApache SupersetMySQLAnalytics
QUEST Alliance · Vocational programmes

An open-source data pipeline and BI dashboard for a non-profit

Siloed programme data brought together with Talend ETL, MySQL and Apache Superset, so vocational programmes report on what is actually happening.

  1. 01The problem Data in silos, and no budget for a proprietary stack.
  2. 02What we built Talend pipelines into MySQL with Superset dashboards on top.
  3. 03What changed Actionable reporting across programmes, on open-source tools.
Open-sourceend to endTalend · MySQL · Supersetthe stackProgrammesreported in one place
In production Read the case study →
Real estate & built environment
JenkinsDockerCI/CDDevOps
Loyalie · Java Spring Boot and Angular

Faster application deployment through DevOps

Manual deployments and release downtime replaced with Jenkins pipelines, Docker containers and code review workflows for a Spring Boot and Angular platform.

  1. 01The problem Every release was a manual event, and a risky one.
  2. 02What we built CI/CD in Jenkins, containerised services, review gates before merge.
  3. 03What changed Deployments without downtime and releases without ceremony.
JenkinsCI/CD pipelinesDockercontainerised platform0release downtime
Delivered Read the case study →
Not sure where to start?

We'll map the highest-value first step for you.

Thirty minutes with an architect who has shipped work like the cards above. You leave with a written first step, whether that is a Decide-phase assessment, a scoped build, or a team.

Book a call → Send a brief instead
What the first call covers
  1. The problem, in your words. What is slow, expensive or blocked, and what it costs today.
  2. The nearest card. Which of the case studies above is closest, and what carried over from it.
  3. A written first step. Scope, people and a number, before anyone builds anything.