An illustrative diagram of your systems in four layers: Experience (Websites & Apps, Search & AI Visibility); AI & automation (AI Strategy & Agents, AI Product & Automation); Software & data (Custom Software & Data Platforms, AI Infrastructure & Cloud, Integration & Support); Trust & operations (Cybersecurity & AI Trust, Audits & Assessments, Tech Workforce). Requests move between the services. A status bar shows a 212-millisecond response time and 148 of 148 AI tests passing. It also shows 71 percent of the outage budget left and 0.18 grams of CO2e per request.
4 shifts drive this: AI in products, monthly model changes, AI answers in search, and stricter security and privacy laws. Each card names its public source.
Screens marked Illustrative are examples, not client results.
rack-14 · in-regiongpu util 64%p95 212 msWhere your software runsYour cloud account and region, monitored at every layer
01A trial→In the product
Then: A trial. Now: In the product.
LLM01
Top risk: AI tricked by hidden text
AI inside the product
AI features now talk to your customers, and they fail in ways normal testing misses.
What we do about it
We test AI on your own examples and add safety checks. A named person approves each release.
Each service builds one or more layers. Pick a layer to see which, or follow one customer question through all 4.
trace 7f3a9c · support.answer
Your systems in 4 layers. Experience: website and web app, mobile app, support chat, search and AI answers. AI and automation: support agent, document search, workflow automation, tests. Software and data: custom apps and CRM, customer data, tool connections, AI gateway, cloud and AI chips. Trust and operations, beside every layer: security and safety checks, live monitoring, audits and the on-call team. The steps below follow one question.
Phases and lengths are typical. We agree scope and price with you before work starts.
Tools we use
The technology we work with, by group.
174 technologies in 18 groups, from languages to team tools. We are not tied to one model or cloud, and pick what tests best.
Technologies we work with. A logo here does not mean we are a partner or reseller, or hold a certificate.
Group
174technologies · 18 groups
In every group, we choose on evidence, write down why and keep each choice reversible.
Use the arrow keys to move between technologies; the selected technology is described in the panel beside the wall.
01Programming languages11
TypeScriptTyped front ends and Node servicesJavaScriptThe browser and everything that runs in itPythonAI, data pipelines and fast back endsGoHigh-throughput services and CLIsRustPerformance-critical components and WebAssemblyKotlinAndroid apps and JVM servicesSwiftiOS and iPadOS appsPHPContent sites and mature web platformsJavaEnterprise back ends on the JVM.NETEnterprise back ends on Windows and LinuxNode.jsAPI servers, workers and tooling
02Website screens9
ReactInteractive product interfacesNext.jsServer-rendered React sites and appsVjVue.jsProgressive interfaces on existing sitesAngularLarge internal applications with strict structureSvelteLightweight embedded widgets and dashboardsAstroContent-heavy sites with minimal JavaScriptTailwind CSSToken-driven styling at speedThree.js3D views and product configurators in the browserWebAssemblyNear-native compute in the browser
03Mobile6
FlutterOne codebase for iOS and AndroidRNReact NativeMobile apps that share logic with React webExpoFaster React Native builds, updates and releasesiOSNative Apple platform featuresAndroidNative Android platform featuresJetpack ComposeModern native Android interfaces
04Server software12
NestJSStructured TypeScript APIsFastAPIPython APIs that serve models and dataDjangoAdmin-heavy platforms with a mature ORMLaravelPHP platforms with a rich ecosystemSpring BootJVM services in enterprise environmentsGraphQLOne typed API over many back endsOpenAPI InitiativeAPI contracts that generate clients and testsRedisCaching, rate limits and queuesRabbitMQReliable work queues between servicesTemporalDurable workflows that survive failuresSupabasePostgres, auth and storage for fast startsFirebaseMobile back ends, push and analytics
05Data & analytics15
PostgreSQLThe default system of recordMongoDBDocument stores for flexible schemasSnowflakeCloud warehouse for analytics at scaleDatabricksLakehouse for data engineering and MLApache SparkDistributed batch and stream processingApache KafkaThe event backbone between systemsApache AirflowScheduled data pipelines with lineagedbdbtTested, versioned transformations in the warehouseClickHouseSub-second analytics on event dataElasticsearchFull-text search and log analyticsGoogle BigQueryServerless warehouse on Google CloudDuckDBFast local analytics inside pipelinesAirbyteConnectors that move data into the warehouseLookerGoverned metrics and dashboardsPBPower BIReporting inside Microsoft estates
06AI & machine learning10
PyTorchTraining and fine-tuning modelsTensorFlowProduction ML in Google-centred stacksHugging FaceOpen models, datasets and tokenisersscikit-learnClassical ML: forecasting, scoring, clusteringNVIDIAGPU compute for training and inferenceMLflowExperiment tracking and model registryvLLMHigh-throughput serving for open modelsRayDistributed training and batch inferenceONNXPortable models across runtimesJupyterExploration and reproducible analysis
07AI models8
OpOpenAITop-tier models behind the gatewayAnthropicTop-tier models for long documents and agent workGoogle GeminiMultimodal models and Google Cloud integrationMistral AIEfficient open-weight and hosted modelsMeta LlamaOpen-weight Llama models run in your cloudDeepSeekOpen-weight reasoning models, self-hostedOllamaLocal models for development and offline usePerplexitySearch-grounded answers and citation tracking
08AI agents, search & testing12
LangChainChains, tools and integrations for LLM appsLangGraphStateful agents with checkpoints and approvalsLlLlamaIndexDocument ingestion and retrieval pipelinesPiPineconeManaged vector search at scaleWeWeaviateHybrid vector and keyword searchQdrantSelf-hosted vector search with filteringMilvusVector search for very large collectionspgpgvectorVectors inside Postgres, no new databaseReplicateHosted open models by APIModalServerless GPUs for batch and bursty inferenceGitHub CopilotAI pair programming in the IDECursorAI-native code editing with review
09Cloud & hosting9
AWAWSPrimary cloud for most regulated workloadsMAMicrosoft AzureCloud for Microsoft-centred estatesGoogle CloudCloud for data- and AI-heavy platformsCloudflareEdge, DNS, WAF and bot managementVercelHosting for Next.js front endsNetlifyHosting for static and Jamstack sitesDigitalOceanSimple cloud for smaller workloadsFastlyEdge caching and computeAkamaiGlobal delivery and edge security
10Build & release tools12
DockerReproducible builds and local environmentsKubernetesOrchestration for services and model servingHelmPackaged Kubernetes deploymentsTerraformInfrastructure as code across cloudsPulumiInfrastructure as code in TypeScript or PythonAnsibleConfiguration for hosts and appliancesArgoGitOps delivery and workflow pipelinesGitHub ActionsAutomatic tests, AI tests and scansGitHubSource, reviews and the audit trailGitLabSource and CI in self-hosted estatesJenkinsCI in established enterprise pipelinesIstioService mesh for mTLS and traffic policy
11Monitoring8
OpenTelemetryTraces, metrics and logs on an open standardPrometheusMetrics and alerting for servicesGrafanaReliability dashboards for engineers and leadersDatadogManaged observability across cloud estatesSentryError tracking in front ends and appsNew RelicAPM in established enterprise stacksElasticLog analytics and security eventsPagerDutyOn-call rotation and incident response
12Testing & quality7
PlPlaywrightEnd-to-end tests in real browsersCypressComponent and end-to-end testsSeleniumCross-browser tests in legacy suitesJestUnit tests for JavaScript and TypeScriptk6Load tests before every launchPostmanAPI contract tests and collectionsSonarQubeCode quality and security gates in CI
13Security & log-in11
OktaEnterprise SSO and identityAuth0Customer identity and accessOpenIDStandard sign-in across systemsJWTSigned tokens between servicesVaultSecrets management and dynamic credentialsSnykDependency and container vulnerability scanningTrivyImage, IaC and SBOM scanning in CIFalcoRuntime threat detection on KubernetesOWASPASVS, Top 10 and LLM Top 10 as the test barBurp SuiteManual penetration testing1PasswordShared secrets for teams, never in chat
14Connections & messaging9
ZapierQuick automations between SaaS toolsMakeVisual automations with branching logicn8nSelf-hosted workflow automation with AI stepsMuMuleSoftEnterprise integration platformKongAPI gateway, auth and rate limitingTwTwilioSMS, voice and verificationWhatsAppCustomer conversations on WhatsApp BusinessSlSlackAlerts, approvals and copilots where teams workMTMicrosoft TeamsApprovals and bots inside Microsoft 365
15Customers, shops & payments11
SaSalesforceCRM integrations and custom objectsHubSpotCRM and marketing automationZohoCRM and operations suites for the mid-marketSAPERP integration for orders, stock and financeShopifyCommerce storefronts and headless buildsWooCommerceCommerce on WordPressStripePayments, billing and subscriptionsRazorpayPayments and UPI for IndiaPayPalGlobal checkout and payoutsIntercomCustomer messaging and support inboxZendeskSupport desk integration and AI handoff
16Search & analytics12
GoogleSearch, Ads and Maps platform APIsGoogle Search ConsoleIndex coverage and query dataGoogle AnalyticsTraffic, conversion and attributionGoogle Tag ManagerConsent-aware tag managementPageSpeed InsightsField and lab Core Web VitalsLighthousePerformance and accessibility audits in CISemrushKeyword and competitor researchAhAhrefsBacklinks and content gapsAlgoliaSite and product searchPostHogProduct analytics, flags and session replayMixpanelProduct analytics and funnelsSoSchema.orgStructured data machines can cite
17Website editors5
WordPressContent sites and editorial teamsContentfulHeadless content across many surfacesSanityStructured content with live previewStrapiSelf-hosted headless CMSWebflowMarketing sites teams edit themselves
18Collaboration & design7
FigmaDesign source of truth and tokensStorybookComponent library and visual testsJiraDelivery tracking in enterprise teamsConfluenceRunbooks and decision recordsLinearFast issue tracking for product squadsNotionDocs and lightweight planningMiroArchitecture and workshop whiteboards
Standards
Frameworks we build to.
We build to these frameworks, so the work produces the proof auditors ask for. The badges are our own drawings. None is an official seal.
Evidence for auditorsEach check is ticked off against what an auditor will ask to see.Photo: Jakub Żerdzicki / Unsplash
Choose a framework to see what it covers, how we use it and which services apply it. Details follow the list.
01Security & privacy14 frameworks
02Safe AI use5 frameworks
03Quality & accessibility3 frameworks
04Sustainability & operations4 frameworks
Without a badge, we also use ISO/IEC 25010 for software quality and the FinOps Framework for cloud costs. NIST AI 600-1 and the NIST AI RMF guide AI risk. Our OWASP Top 10 for LLM Applications entry follows the 2025 list.
"Built to" describes how we work. It does not mean Xterra Edze holds a certificate. ISO 9001 and ISO 14001 cover whole organisations; we follow their practices. Certificates Xterra Edze holds: to be confirmed before launch.
Framework1 / 26
Security & privacy · Standard
ISO/IEC 27001:2022
Information security management systems · ISO / IEC
What it covers
Rules for managing information security risks, with 93 controls on organisation, people, buildings and technology.
How it shows up in our work
As we work, we review access, approve changes, check suppliers and keep logs for your auditor.
Security & privacy · Attestation framework
SOC 2
Trust Services Criteria · AICPA
What it covers
Controls for security, staying online, correct processing, confidentiality and privacy. Type I checks the design on one date. Type II checks it works over time.
How it shows up in our work
We record change approvals, access and incidents automatically, ready for your auditor.
Security & privacy · Framework
NIST CSF 2.0
Cybersecurity Framework · NIST
What it covers
Managing security risk in 6 parts: Govern, Identify, Protect, Detect, Respond and Recover.
How it shows up in our work
We compare your security with your target, then rank and cost the steps.
Security & privacy · Framework
CIS Controls v8.1
Critical Security Controls and Benchmarks · Center for Internet Security
What it covers
Security steps in order of priority, plus secure settings for cloud, containers and operating systems.
How it shows up in our work
We set up cloud accounts and servers to CIS Benchmarks and watch for unapproved changes.
Security & privacy · Standard
OWASP ASVS
Application Security Verification Standard · OWASP
What it covers
Security checks for web apps and links between tools, in 3 risk levels.
How it shows up in our work
We pick the level your risk needs, add its checks to each task and test them automatically.
Security & privacy · Awareness standard
OWASP Top 10
Web application security risks · OWASP
What it covers
The top website security risks, such as weak access control, harmful input and bad settings.
How it shows up in our work
We plan for attacks early, scan each change and test by hand before release.
Security & privacy · Framework
SLSA
Supply-chain Levels for Software Artifacts · OpenSSF
What it covers
Levels of proof that software was built safely, with a checkable build record.
How it shows up in our work
We sign each release, with a build record and parts list (SBOM) in SPDX or CycloneDX.
Security & privacy · Standard
PCI DSS v4.0.1
Payment Card Industry Data Security Standard · PCI Security Standards Council
What it covers
12 rules to protect card data wherever it is stored, used or sent.
How it shows up in our work
We limit where card data goes, with stand-in numbers and separate networks, then protect the rest.
Security & privacy · Regulation
HIPAA Security Rule
Safeguards for electronic protected health information · US HHS
What it covers
Policy, physical and technical safeguards for patients' electronic health data (ePHI).
How it shows up in our work
We limit access, record every action and encrypt health data. Each supplier signs a HIPAA agreement.
Security & privacy · Regulation
GDPR
General Data Protection Regulation (EU) 2016/679 · European Union
What it covers
Europe's data law: a legal reason to use data, people's rights, privacy by design, breach reports and risk checks.
How it shows up in our work
We build consent, time limits on data and people's requests into the product, with risk checks where needed.
Security & privacy · Law
DPDP Act 2023
Digital Personal Data Protection Act, 2023 · Government of India
What it covers
India's data law and its Rules: notice and consent, people's rights, breach reports and extra duties for significant firms.
How it shows up in our work
From the first release, we build consent records, notices and a way to handle requests from users in India.
Security & privacy · Standard
ISO/IEC 27701
Privacy information management systems · ISO / IEC
What it covers
Rules for handling personal data you collect, or process for someone else.
How it shows up in our work
We record where personal data is kept, for how long and who handles it.
Security & privacy · Regulation
CERT-In Directions 2022
Cyber incident reporting directions · CERT-In, Government of India
What it covers
India's cyber-security rules: report listed incidents in 6 hours, keep logs in India for 180 days and sync clocks.
How it shows up in our work
We set up incident steps, log storage and synced clocks to meet the deadlines.
Security & privacy · Directive
NIS2
Network and Information Security Directive (EU) 2022/2555 · European Union
What it covers
Europe's cyber-security rules for key organisations: risk measures, supplier security and staged incident reports.
How it shows up in our work
For clients it covers, we fit its risk and reporting steps into what we build and run.
Safe AI use · Standard
ISO/IEC 42001:2023
Artificial intelligence management systems · ISO / IEC
What it covers
Rules for managing AI in a company: policy, impact checks and controls on data and models.
How it shows up in our work
Before an AI tool goes live, we list it, check its impact and set its controls.
Safe AI use · Framework
NIST AI RMF 1.0
AI Risk Management Framework · NIST
What it covers
Managing AI risk in 4 parts: Govern, Map, Measure and Manage. A second guide covers generative AI (NIST AI 600-1).
How it shows up in our work
We test each AI job's risks on your own examples. A named owner approves each release.
Safe AI use · Regulation
EU AI Act
Artificial Intelligence Act (EU) 2024/1689 · European Union
What it covers
Europe's AI law sets rules by risk: banned uses, duties for high-risk AI and general-purpose models, and AI labels.
How it shows up in our work
We rate each AI job by risk early, and build in AI labels, records and human checks.
Safe AI use · Awareness standard
OWASP Top 10 for LLM Applications
Security risks in generative AI applications · OWASP GenAI Security Project
What it covers
Risks in AI apps, such as hidden instructions that trick the AI, data leaks and AI with too much power.
How it shows up in our work
Before any AI goes live, we attack it with tests for hidden instructions, data leaks and tool misuse.
Safe AI use · Knowledge base
MITRE ATLAS
Adversarial Threat Landscape for AI Systems · MITRE
What it covers
Known ways attackers target AI, collected from past attacks and research.
How it shows up in our work
We list likely attacks on your AI from ATLAS, then try them in tests.
Quality & accessibility · Standard
WCAG 2.2 AA
Web Content Accessibility Guidelines · W3C
What it covers
Rules that make websites usable by people with disabilities. Version 2.2 adds visible focus, tap size and easier log-ins.
How it shows up in our work
We check each change automatically and test main tasks by keyboard and screen reader.
Quality & accessibility · Metric set
Core Web Vitals
Loading, interactivity and visual stability · Google
What it covers
For 3 in 4 visits, pages show in 2.5 s, respond in 0.2 s and do not jump around.
How it shows up in our work
We check speed limits on every change and track visitors' speed on each page type.
Quality & accessibility · Standard
ISO 9001:2015
Quality management systems · ISO
What it covers
Rules for managing quality: clear processes, attention to risk and steady improvement.
How it shows up in our work
We agree what done means, review each other's work and use release checklists.
Sustainability & operations · Specification
SCI · ISO/IEC 21031:2024
Software Carbon Intensity · Green Software Foundation
What it covers
The carbon software causes per user or other unit, counting energy use and the making of hardware.
How it shows up in our work
We measure carbon per request, user or AI answer as a starting point.
Sustainability & operations · Metric set
DORA metrics
Software delivery performance · DORA (DevOps Research and Assessment)
What it covers
How often you release, how fast changes go live, failure rate and recovery time.
How it shows up in our work
We measure them in your release tools from week one and report them as numbers.
Sustainability & operations · Standard
ISO 14001:2015
Environmental management systems · ISO
What it covers
Rules for managing environmental impact, legal duties and steady improvement.
How it shows up in our work
We track hosting, device and data impact beside cost, with targets to cut it.
Sustainability & operations · Standard
ISO 22301:2019
Business continuity management systems · ISO
What it covers
Rules for keeping your business running in a crisis: plan, test and improve.
How it shows up in our work
We set recovery time and data-loss limits for each service, then test them in drills.
A live activity log. Each line shows who acted, what they did, on what, how they signed in, the result and a trace ID. It includes an AI agent using a tool, a blocked export and a person approving a release.
retention 400 d · in-regiontamper-evident hash chainclock NTP · IST/UTC
How we use AI
AI helps build your software, under 4 rules.
AI agents draft the plan, tests and release. Each change passes code tests, AI tests and security scans. Below, one task runs through 8 stages.
Reviewed by peopleAI drafts code; an engineer reads every line before it goes in.Photo: Compagnons / Unsplash
An interactive build console. One task, adding refund status to a support assistant, runs through eight stages. An agent drafts the plan and the tech lead approves it. An engineer writes the code with an AI helper and submits change 1482. A test agent adds 14 tests. The AI tests pass 148 of 148 examples, with a true-to-sources score of 0.94. Security scans find nothing critical. A named engineer reviews and approves the change. An agent releases it to 5, then 100 percent of users. The live system is then monitored against its targets. Other tabs show the code changes, the AI test report and the activity log.
delivery / your-platform / XE-1482 · Refund status in the support assistantLive after 3 h 32 min
Stage 01 of 8Agent + person
Plan
From the ticket, an agent drafts how we agree it is done and the edge cases. The tech lead edits and approves.
Output6 criteria · 3 edge cases · approved
Stage 02 of 8Agent + person
Code
An engineer writes the change with an AI helper in the code editor, then submits it with a summary.
OutputPR #1482 · +212 −38 · 7 files
Stage 03 of 8Agent
Tests
A test agent writes automatic tests for the change. The engineer keeps the ones that earn their place.
Output+14 tests · coverage 81.2% → 84.6%
Stage 04 of 8Agent
Checks
The AI feature is tested on your own examples: sources, refusals, speed and cost, against the live version.
Output148 / 148 pass · true to sources 0.94
Stage 05 of 8Agent
Security
Scans check the code, its add-ons and any stored passwords, with hidden-instruction attacks (OWASP LLM01) included.
Output0 critical · 0 high · 0 secrets
Stage 06 of 8Person
Review
A named engineer reads the change, the AI test report and the scans, then approves or asks for changes. Agents cannot approve.
!Waiting for a named engineer. Agents cannot approve.
✓Approved by the tech lead · 17:12 · logged
Stage 07 of 8Agent
Release
Goes live to 5% of users, then 25% and 100% while errors and speed hold. If they slip, it undoes itself.
Outputrollout 5% → 25% → 100% · none undone
Stage 08 of 8Agent + person
Monitor
Speed, reliability and AI quality are watched once live. The on-call engineer handles anything that slips.
OutputOnline 99.96% · on target · AI steady
run.logXE-1482 · run 3
14:02spec.agent drafted 6 acceptance criteria, 3 edge cases · approved by tech lead
14:31ide.pair PR #1482 opened · +212 −38 · 31 of 38 suggested lines kept
16:40test.agent +14 tests (12 kept) · coverage 81.2% → 84.6% · all green
16:46eval.runner golden-set@v14 148/148 · faithfulness 0.94 · p95 1.9 s
Why it mattersModels and instructions change. The same 148 tests run on every change, so a drop in quality is stopped before customers see it.
Time
Actor
Action
Model · prompt
Reviewer
Result
14:02
spec.agent
Drafted acceptance criteria
route:large · spec@v7
Tech lead
Approved
14:31
ide.pair
Suggested 38 lines in 3 files
route:code
Tech lead
31 kept
16:40
test.agent
Generated 14 tests
route:code · tests@v3
Tech lead
12 kept
16:46
eval.runner
Ran golden set v14
judge@v5
—
Pass
17:12
Tech lead
Approved merge of PR #1482
—
—
Merged
17:34
deploy.agent
Canary 5% → 100%
—
On-call
Live
RetentionKept with each change and release, exportable to your security system and sampled in the monthly audit.
Agents propose, people approve
Agents draft, test and release. A named engineer approves every change before it goes live.
Every AI action is logged
The model, instructions, inputs, output and reviewer are kept with each change and release.
No client data in public training
Business AI accounts with training switched off, in your region where required.
Passwords never reach the AI
Passwords stay in a secure vault. Agents get limited, short-lived access.
What we track on every project · DORA metrics Illustrative
Time to go live
Measured in hoursTask to live for the run above: 3 h 32 min
How often we release
Daily when neededSmall, reversible changes, released in stages
Changes that fail
Tracked per releaseEvery undo is reviewed in writing
Recovery time
Tracked per incidentGuides and on-call from day one
Model updates
New models, tested within days of release.
We test each new model on your examples and switch only if it scores better or costs less. Changes go live weekly, after 4 checks.
Model log · last 12 months
9 new models, each tested on your examples
Test score
0.81 → 0.93
Cost per 1k requests
$2.40 → $1.02
Adopted
6 of 9
Median time to decision
2 days
A chart of the last twelve months. Nine model releases are marked along the top. The test score on your examples rises in steps from 0.81 to 0.93 as six are adopted. Over the same time, the cost per thousand requests falls from 2.40 to 1.02 dollars. Each release can be chosen to see how it was tested and what changed.
Test score · your examplesCost per 1k requestsAdoptedTested, not adoptedIllustrative
Weekly releases · 12 weeks Illustrative
Every Thursday, whatever passed the checks goes live
12 weekly releases. 11 went live, with 9 to 23 changes each. Week 7 was held because a refund test score fell from 0.92 to 0.86. The fix went live in week 8.
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
W11
W12
✓Testsparts · links · journeys
✓AI testsyour examples · every AI feature
✓Scanscode · add-ons · passwords
✓Rollout5% → 100% while targets hold
!W7 held. A refund test score fell from 0.92 to 0.86. Nothing went live; the fix went out with W8.
11 of 12 releases went live1 held by AI tests0 rollbacks
Idea to agent · typical Typical ranges
From a clear problem to live use in about 6 weeks
T+Week 6
Day 0Problem framedOne task, one owner and one measure, plus how the work is done today.Owner signs the measure
Day 2Test version on real dataA working agent on anonymised copies of your data, logged step by step.The team who does the work reviews outputs
Day 5Starting scoreA test set from your real cases, scored for quality, cost and speed.Go or stop, decided on the numbers
Week 3Trial with usersNamed users; a person approves every action; safety checks and spend limits on.Security and legal sign the safety rules
Week 6Live, with safety checksReleased through the AI gateway with monitoring, guides and an activity log.Sponsor signs the release
How a project runs
From first call to software your teams run.
Product, engineering, AI and security people work in shared 2-week cycles. 4 checks decide when work moves on. Pick a phase to see its decision, team and AI tasks.
A 5-phase project plan. Discover: weeks 1 to 2. Design: weeks 2 to 4. Build: weeks 4 to 16. Launch: weeks 16 to 18. Run: from week 18. 4 lanes: product and design; engineering; AI and data; security and quality. Checks: system design approved after Design, security sign-off after Build, go-live after Launch and a 90-day review in Run.
Interviews · journey mapPrototype tested with usersDesign in every cycle · design kitTraining · launch commsUsage data · next plans
Engineering
Map of systemsDesign · decision logTwo-week cycles · demo each cycleSwitch-over · supportRun for you · targets
AI & data
Data & readiness auditTest examples · test planDoc search, agents · auto testsTrial → rolloutQuality watched once live
Security & quality
Likely attacks · privacy checkCode scans · tests · attack testsSecurity · load testsFix times for flaws · audit
Phase 01 · Wk 1–2
Discover
Ends in a decisionWhat to build first, what it must achieve and whether the data can support it.
What you get
The problem and success measures
A map of your systems today
Data and AI readiness check
Risk list and privacy check
Who takes part
Your sponsorYour product ownerDomain expertsProject leadSolution architectAI lead
Where AI helps
Turns interview notes and tickets into themes, with sources
Maps your systems and data flows from documents and code
Drafts a first risk list for people to review
Phase 02 · Wk 2–4
Design
Ends in a decisionThe system design, the build order and the checks the build must pass.
Check · System design approved
What you get
System design and a record of decisions
Prototype tested with users
AI test examples, test plan
Security design and likely attacks
Cycle plan with goals
Who takes part
Solution architectTech leadProduct designerSecurity engineerYour IT and security leads
Where AI helps
Drafts design options and trade-offs for the architect to decide
Makes draft screens from your shared design kit
Suggests AI test examples from past tickets
Phase 03 · Wk 4–16
Build
Ends in a decisionEvery 2 weeks: what this cycle delivers, shown working in a demo.
Check · Security sign-off
What you get
Working software every cycle
Code tests, AI tests and scans on every change
Guides written as features land
A decision log kept current
Who takes part
Tech leadEngineersProduct designerTest and AI-test engineerYour product owner, weekly
Where AI helps
Helps engineers write code; people review every line
Writes tests and change summaries
Runs AI tests on every change and flags anything worse
Phase 04 · Wk 16–18
Launch
Ends in a decisionGo live, with an undo plan rehearsed and on-call staff ready.
Check · Go-live
What you get
Peak-traffic test report
Independent security test report
Switch-over plan and undo rehearsal
On-call rota, guides and training
Who takes part
Release managerReliability engineerSecurity engineerYour support leadYour sponsor
Where AI helps
Checks AI answers against your team's decisions in a pre-launch trial
Watches the step-by-step release and undoes it if errors rise
Drafts release notes and training material
Phase 05 · Wk 18 →
Run
Ends in a decisionEvery month: what to improve next, decided on service targets, test results and cost.
Check · 90-day review
What you get
Monthly report: service targets, AI test results, cost and carbon
Incident reviews without blame
Ranked list of improvements
The 90-day review
Who takes part
Service ownerOn-call teamYour product ownerYour finance partner, for cloud cost
Where AI helps
Sorts alerts and drafts incident timelines
Watches AI test scores after model or data changes
Flags unusual costs and suggests a fix for approval
Design · week 3Design options on the wall before any are built
2-week cycles
A goal per cycle, agreed with your product owner.
A demo every cycle
Working software you can click through on a test site.
A decision log
Each design choice written down, so it outlives the people who made it.
Guides before go-live
Written as features land and rehearsed before launch.
Sustainability
Lower-carbon software, measured per request.
AI adds energy to every request. We measure carbon per request with the Software Carbon Intensity (SCI) standard, ISO/IEC 21031. What cuts carbon also cuts cost.
Measured and reportedCarbon reduced by design, then measured
01
Reported per unit of use
SCI per request or per active user, with what is counted written down.
02
No offsets in the figure
The SCI standard leaves out certificates and offsets, so the figure counts only cuts we make.
03
Choosing the cloud region
We weigh speed, grid carbon and data location rules under the DPDP Act. We record the choice.
0.19 Wh per request, including data-centre overhead
I Grid intensity
653 gCO₂e per kWh where it runs
M Embodied
0.053 g, the hardware's share of its own manufacture
R Functional unit
One answered request
Region · grid intensity and round trip from Delhi users
India West is the default because it carries the widest service coverage of the Indian regions, which usually decides it. India North is cleaner and closer to Delhi users, so it wins wherever its services cover the workload. The choice is made per project; both keep data in India.
Practices
Where the carbon went · gCO₂e per request
Baseline0.47
Right-sized models−0.18
Fewer AI calls−0.05
Timed for cleaner power−0.01
Lighter pages−0.01
Idle servers switched off−0.04
SCI now0.18
0.18gCO₂e per request
62% below baseline · 179 g per 1,000 requests
Right-sized models
A small model answers first and hands over to a large one only when unsure.
Energy per request down 30–60% where used
Fewer AI calls
Repeat questions get a saved answer; background jobs run in batches.
Model calls down 20–40%
Timed for cleaner power
Heavy AI jobs and reports run when and where power is cleaner, using live grid data.
Job emissions down 10–30%
Lighter pages
Every build checks size limits for images, fonts and code, so each visit sends less data.
Page weight down 30–50%
Idle servers switched off
Servers scale to zero overnight, sizes match the load and unused resources are deleted monthly.
Idle spend and hardware carbon down
Typical effects are published ranges. Grid figures are illustrative yearly averages; live ones come from your cloud and grid data.
Industries
Sector rules built into the software.
Each sector has its own laws, rules on where data lives, speed needs and human review. We settle them before coding starts. Pick yours.
Sector01 / 08
These notes summarise the rules. They are not legal advice.
99% of payment calls ≤ 180 ms
01 · PCI DSS · RBI · SEBI
Financial services
Payments must be fast, and every decision needs a record.
Regulation
PCI DSS v4.0.1 for card data, RBI rules on IT outsourcing and SEBI's cybersecurity and cyber-resilience framework for market firms.
Where data lives
Payment data stays in India under RBI's 2018 direction; customer data follows the DPDP Act.
Speed & scale
Payment and trading systems answer 99% of requests in a few hundred milliseconds, even at salary-day and market-open peaks.
AI oversight
Every AI-assisted decision is logged with its inputs, model and reviewer. Credit decisions come with reasons.
ISO/IEC 27001:2022 — Information security management systems
GDPR — General Data Protection Regulation (EU) 2016/679
EU AI Act — Artificial Intelligence Act (EU) 2024/1689
Typical first projectSecurity-review readiness, then an AI feature that keeps client data apart.
Our reports
A monthly report in numbers everyone reads the same way.
We report speed, reliability, AI quality, security, cost and carbon each month. Each number has a written definition and a target agreed before launch.
Service report · Xterra Edze
Your systems
Six measures · targets agreed before launch · a definition by every number Illustrative
01SpeedOn target
Load time · mobile, real visits
2.1s▼ 1.7 s · 12 months
Good ≤ 2.5 s
12 months agoNow
p95 response time212 ms
Definition · Speed
How long the main content takes to appear, for 75% of real page loads on mobile (LCP). Google's Core Web Vitals rate 2.5 s or less as good. p95 response time: 95% of requests finish faster than this.
02ReliabilityOn target
Uptime vs target · last 30 days
99.97%▲ 0.25 pts · 12 months
Goal ≥ 99.9%
12 months agoNow
Outage budget left71% · 12.5 of 43.2 min used
Definition · Reliability
The share of requests served well over 30 days. A 99.9% target allows 43.2 minutes of failure a month: the outage budget. When it runs out, releases pause and reliability work comes first.
03AI qualityOn target
True to sources · test set
0.94▲ 0.10 · 12 months
Pass ≥ 0.90
12 months agoNow
AI tests passed148 / 148 cases
Definition · AI quality
The share of claims in an answer that its sources support. We call it true to sources. A judge model, checked against expert ratings, scores it. The same set of your real examples runs on every change to a model, instruction or data source. The hero readout, the build console and the planning demo all report this one figure.
04SecurityOn target
Open critical security flaws
0▼ 6 · 12 months
Target 0 overdue
12 months agoNow
Fix time · high severity3.2 days
Definition · Security
Confirmed critical flaws from scans, attack tests and reports, counted until the fix is checked live. Fix time is the average time from a flaw being confirmed to its fix going live.
05CostOn target
Cost per 1,000 requests
$1.02▼ $1.38 · 12 months
Cap ≤ $1.20
12 months agoNow
Spend vs monthly budget94%
Definition · Cost
Servers, AI usage, storage and data transfer, divided by requests served, per 1,000. Reviewed with finance every month using the FinOps Framework, because cost per request says more than the total bill.
06CarbonOn target
SCI per request
0.18gCO₂e▼ 0.29 g · 12 months
Target ≤ 0.25
12 months agoNow
Jobs run at cleaner hours86%
Definition · Carbon
Software Carbon Intensity from the Green Software Foundation (ISO/IEC 21031:2024): energy, grid carbon and hardware, per request. Offsets are left out, so it falls only when the software uses less.
Weekly
Operations review
Outage budget used, incidents, AI test scores and any call-outs, in 15 minutes with the on-call engineer.
Monthly
Service report
All 6 measures against target, including cost and carbon, and next month's changes.
Quarterly
Business review
The measures agreed before launch, the plan ahead and our delivery numbers (DORA).
Core Web Vitals (speed)
Loads in 2.5 s, reacts in 200 ms and barely shifts, for 75% of visits
Outage budget
99.9% over 30 days = 43.2 min of allowed failure
Early warning
1 h at 14.4× the safe rate uses 2% of the budget: alert on-call
p95 / p99
The time 95% or 99% of requests beat; averages hide the slowest ones
Where we work
Offices in India and Canada, overlapping your hours.
Together, our office hours overlap most of the working day in Europe, the Gulf, South-East Asia and North America.
Time zones
2India Standard Time (UTC+5:30) and Atlantic Time in Saint John (UTC−4, UTC−3 in summer)
Office hours
10:00–19:009 hours in India, Monday to Friday
Extended shift
+ 3.5 hTo 22:30 IST, by agreement
On call
24 / 7Rota and who to call, by agreement
Office 01 · India
New Delhi
00:09 IST · UTC+5:30
6 Worldmark, AerocityNew Delhi 110037, India
Work led here
Client and project leadership
System design and AI engineering
Security, audit and assessment work
Office 02 · India
Ludhiana
00:09 IST · UTC+5:30
SCO-2, LGFSCO-1, 3rd FloorNoble Enclave, Opp. Hotel Park PlazaFerozepur Road, Ludhiana, Punjab 141001, India
Work led here
Cloud, product and web engineering
Data, testing and AI checks
Managed service, support and the on-call rota
Office 03 · Canada
Saint John
14:39 AST · UTC−4
120 University AvenueSaint John, NB E2K 1Z3, Canada
Overlap
Your working day against ours, hour by hour
Hours in India000306091215182124Shared
Offices · IndiaIST · UTC+5:30
H1H2
9 hoffice hours
Office · CanadaSaint John · ADT · UTC−3Saint John · AST · UTC−4
8 hoffice hoursSaint John office, Northern summer: open 09:00–17:00 ADT, which is 17:30–01:30 India time.Saint John office, Northern winter: open 09:00–17:00 AST, which is 18:30–02:30 India time.
United KingdomLondon · BST · UTC+1London · GMT · UTC+0
8.5 hin our office hoursUnited Kingdom, Northern summer: a 09:00–17:30 local working day shares 8.5 h with our office hours.8.5 hin our office hoursUnited Kingdom, Northern winter: a 09:00–17:30 local working day shares 8.5 h with our office hours.
EuropeFrankfurt · CEST · UTC+2Frankfurt · CET · UTC+1
8.5 hin our office hoursEurope, Northern summer: a 09:00–17:30 local working day shares 8.5 h with our office hours.8.5 hin our office hoursEurope, Northern winter: a 09:00–17:30 local working day shares 8.5 h with our office hours.
GulfDubai · GST · UTC+4Dubai · GST · UTC+4
9 hin our office hoursGulf, Northern summer: a 09:00–18:00 local working day shares 9 h with our office hours.9 hin our office hoursGulf, Northern winter: a 09:00–18:00 local working day shares 9 h with our office hours.
5.5 hin our office hoursSingapore, Northern summer: a 09:00–18:00 local working day shares 5.5 h with our office hours.5.5 hin our office hoursSingapore, Northern winter: a 09:00–18:00 local working day shares 5.5 h with our office hours.
US EastNew York · EDT · UTC−4New York · EST · UTC−5
7 hin our office hoursUS East, Northern summer: a 09:00–17:30 local working day shares 7 h with our office hours.7 hin our office hoursUS East, Northern winter: a 09:00–17:30 local working day shares 7 h with our office hours.
Now 00:09
Our office hoursExtended shift, by agreementOn-call rota, by agreementShared hoursH1 · H2 handovers
Working days shown as 09:00–17:30 local (09:00–18:00 in the Gulf and Singapore). Shared hours count office days in India and Saint John. Handovers are written down: open incidents, releases in progress and pending approvals.
Offers & plans
Technology services, bought alone or as one project.
Pick an offer to start with and your enquiry reaches the right team. Each comes as a short Sprint, a fixed-price Project or an ongoing team.
01Pick offers. Ask about one, or add several to one enquiry.
02Choose a plan. A small job, a fixed project or ongoing help.
03Send your enquiry. We reply within one working day.
How we work with you
Ways to start, from a question to a quote.
Ask a quick question, send your project details or request a quote. The person who will lead the work reads each one in full. Services in your enquiry go with it.
For a first chat, a press request or anything not ready to plan yet.
You getA reply from a named lead
Start
02Recommended
About 8 minutes5 short steps
Goals, who it is for, budget and timing, so our first reply can outline the work.
You getOptions and prices after one call
Start
03
About 15 minutesYour documents attached
Your documents, deadlines and the purchasing and security rules to meet.
You getConfirmation and a named lead
Start
How it is priced06
Each plan shows how it is priced. Work starts once we agree a written scope and price.
Opens the project form with this plan chosen.
01In your enquiry
Typical length
1–3 weeks
Pricing
Fixed fee
02In your enquiry
Typical length
4–12 weeks
Pricing
Fixed price
03In your enquiry
Typical length
3–9 months
Pricing
Fixed price per stage
04In your enquiry
Typical length
Ongoing · 6-month minimum
Pricing
Monthly fee
05In your enquiry
Typical length
6–18 months
Pricing
Priced per written scope
06In your enquiry
Typical length
Ongoing · 3-month minimum
Pricing
Paid by time used
Compare what each plan includes
What each plan includes and who it suits
Plan
What you get
Best for
SprintA short, fixed job that answers one clear question.
Scope and result agreed before we start
A senior lead and the right specialists
A weekly review with you
A clear answer or a test version
An audit, a brand check or a test version of an app
ProjectOne clear job for one fixed price.
A written scope: what you get and how we agree it is done
A named project lead and a fixed team
A shared plan with checkpoint dates
Source files, yours once paid for
A brand identity, a website, a simple app or a set of tools
MilestoneA bigger build in stages you approve and pay for one at a time.
Stages, each with its own scope and approval
A go or stop review at each stage
Re-plan between stages as you learn
Pay for each stage you approve
Work too big for one contract, such as a large app
RetainerA monthly team that runs and improves what we built.
Team time set aside for you every month
Agreed response times for requests and fixes
A monthly review and a to-do list
Planned improvements, not only fixes
Monthly social media and ads, or care for live sites and apps
EnterpriseA large, long project with its own team and agreed response times.
A project director and a decision group
A team for each part of the work
Response times, reports and a list of risks
Security, legal and purchasing checks in the plan
Large companies making changes across markets, brands or business units
SquadOur specialists join your team and work in your tools.
Named specialists matched to your plan
Works in your own tools and meetings
Grow or shrink the team each month
We teach your team from week one
Teams with a clear plan that need senior people fast
How we work with you
6 terms we build into the work.
The answers your legal, IT, security and buying teams ask for, each with the setting that enforces it.
Reviewed togetherDecisions written down as they are made
terms.md · Your company × Xterra Edze6 clauses
§1Usually asked by Legal · Procurement
You own the code and the AI settings
The code, cloud set-up, AI instructions, AI test examples and model settings sit in your code accounts from day one. All of it is yours once it is paid for.
Every model sits behind our AI gateway and your test examples. Switching provider, or running a model yourself, is a settings change we can show you, not a rebuild.
Enforced asgateway config
route "support.answer" { model = var.model }switch.requires = "golden set ≥ gate"exit = open formats, your keys
§3Usually asked by Security · IT
Secure from the first line of code
Only the right people get access, with your one log-in where you have it. Passwords sit in a vault, every build is scanned and your security questions are answered before work starts.
Enforced asCI policy
ci.required = [sast, deps, secrets, iac, image]access = sso + mfa, least privilegesecrets = vault only · never in prompts
§4Usually asked by Security · Risk
People approve, agents assist
Agents draft, test and suggest. A named person approves every code change, every live change and anything that reaches your customers. Every AI action is logged.
Enforced asbranch protection
main.required_reviews = 1 # a named personagents.can_merge = falseai_actions.log = [model, prompt, reviewer]
§5Usually asked by CTO · Finance
Measured, not claimed
Targets for speed, reliability, AI quality, security, cost and carbon are agreed before launch. We report them every month, good months and bad.
Design decisions, step-by-step guides, dashboards and training notes are written as the work lands. You sign them off like any other part, so your team can run it without us.
Whichever scores best on your own examples. We use OpenAI, Anthropic, Google, Meta and Mistral models, and open models in your cloud. We pick per task on quality, speed and cost, and switch only when a new model does better.
Yes. It runs in your AWS, Azure or Google Cloud account, India regions included, with training on your data switched off. Where data cannot leave your network, open models run inside it.
You do: code, AI instructions, test examples and model settings. What we make for you is yours once it is paid for. Tools we already had stay ours, and you get a free, permanent licence to use them.
Early and in writing. We answer your security questions before work starts and follow your policies. The proof auditors ask for comes with the work. Working towards ISO/IEC 27001 or SOC 2? We prepare you for the independent auditor.
Typically 4–8 weeks from a clear problem to live use. A test version runs on your data within days. A trial with named users starts by week 3. Regulated or high-risk uses take longer to review.
You choose. We run it with agreed response times, or hand it to your team with guides, dashboards and training. Either way, we review the first 90 days against the agreed measures.