AI and cloud engineering for products that ship, scale and last.

Nagami Ltd is a UK-based consultancy that designs, builds and operates AI-powered products on AWS and Google Cloud — with deep, hands-on expertise in Anthropic's Claude models, from retrieval-augmented assistants to agentic workflows and Claude Code enablement.

Cloud-native deliveryAWS & Google Cloud architecture
Applied AI engineeringClaude assistants, RAG & agents
Production operationsMLOps, LLMOps & FinOps
About Nagami

A senior engineering consultancy, built for the AI era

Nagami Ltd is a private limited company registered in England and Wales. We are a hands-on engineering consultancy: our consultants are practising cloud architects and software engineers who design systems, write production code and stay accountable for what we ship. We specialise in cloud-native platforms on Amazon Web Services and Google Cloud, and in building generative AI products on Anthropic's Claude models — including retrieval-augmented generation, evaluation-driven development and agentic automation.

We work with startups, scale-ups and established product companies across EMEA and Asia — teams that need to move quickly without accumulating technical or compliance debt. Engagements range from a two-week architecture review to the end-to-end delivery and ongoing operation of an AI-powered product. Whatever the size, the standard is the same: secure by default, measured against agreed outcomes and documented so your team can own it after we leave.

Services

What we do

Six focused practices, one delivery standard. Most engagements combine two or three of these — for example, a Claude-based assistant delivered on a well-architected AWS foundation with evaluation and cost controls built in from day one.

Cloud Architecture & Migration (AWS)

Well-architected reviews, landing zones and migrations to AWS. We design for resilience and cost from the start — EKS, Lambda, Aurora, and infrastructure as code with Terraform and CDK.

Data & Analytics Platforms (Google Cloud)

Modern data platforms on Google Cloud: BigQuery warehouses, streaming pipelines with Pub/Sub and Dataflow, dbt transformation layers and governed self-service analytics.

Generative AI & LLM Integration (Claude)

Production systems on Anthropic's Claude models: retrieval-augmented generation, tool use and agents, prompt architecture, guardrails and rigorous offline and online evaluation.

MLOps & LLMOps

The unglamorous work that makes AI dependable: CI/CD for models and prompts, evaluation harnesses, drift and quality monitoring, versioned datasets and safe rollout strategies.

Managed Cloud Operations & FinOps

Ongoing operation of the platforms we build: 24/7-ready observability, incident response, security patching and continuous cost optimisation with clear monthly reporting.

AI Product Engineering

Full-stack delivery of AI-first products: API design, web and mobile front ends, event-driven backends and developer enablement — including Claude Code adoption programmes for engineering teams.

Why Nagami

How we work

Senior engineers only

Every engagement is staffed by experienced practitioners — no delivery pyramids, no hand-offs to a bench. The people who scope the work are the people who build it.

Security first

Least-privilege access, encrypted data paths, auditable pipelines and privacy-by-design for AI workloads. We treat security and compliance as design inputs, not a final checklist.

Outcome-based engagements

We agree measurable targets up front — latency, cost, accuracy, time-to-market — and report against them throughout. You always know what you are getting and why.

Case Studies

Selected work

A representative sample of recent engagements. Client names are withheld under confidentiality agreements; each client is described by industry and scale.

Re-platforming a global language-learning app onto AWS

AWS · Cloud Architecture

Client: a language-learning mobile app with over 10 million registered users across Asia and Europe.

The challenge

The client's messaging and content services ran on a fixed fleet of virtual machines that had grown organically over eight years. Traffic peaks around school terms and marketing campaigns forced permanent over-provisioning, release cycles took weeks, and a regional outage would have interrupted service for millions of daily conversations.

The solution

We led a phased migration to a containerised platform on Amazon EKS with Karpenter for just-in-time node provisioning, moved the primary datastore to Amazon Aurora MySQL with read replicas in a second region, and introduced Amazon ElastiCache for Redis for session and presence data. Media delivery moved behind Amazon CloudFront with S3 origins, and the whole estate was rebuilt as Terraform modules with GitHub Actions pipelines and canary deployments. CloudWatch and Grafana dashboards now give the client per-service cost and latency visibility.

Amazon EKSKarpenterAurora MySQLElastiCache for RedisCloudFrontAmazon S3TerraformGitHub Actions
34%reduction in monthly compute spend after right-sizing and autoscaling
410ms → 170msp95 API latency for chat message delivery
Weekly → dailyrelease cadence, with automated canary rollbacks

A unified analytics platform for a European marketplace

Google Cloud · Data

Client: a European e-commerce marketplace processing several million orders per year across five countries.

The challenge

Order, catalogue and clickstream data lived in a dozen disconnected systems. Finance closed the month from spreadsheets, merchandising decisions relied on day-old exports, and every new report required engineering time. The leadership team wanted a single governed source of truth — without hiring a large data department.

The solution

We built a lakehouse on Google Cloud: change-data-capture from operational databases via Datastream, event ingestion through Pub/Sub and Dataflow into BigQuery, and a documented dbt transformation layer with tested, versioned models. Looker Studio dashboards serve the business teams, while data contracts and Dataplex-based lineage keep the platform governed. Cost controls include partitioned tables, slot reservations and per-team usage reporting.

BigQueryDataflowPub/SubDatastreamdbtDataplexLooker StudioCloud Composer
24h → 20mindata freshness for core sales and stock reporting
5 days → 1 daymonthly finance close preparation time
41%lower cost per query versus the legacy warehouse, at triple the usage

A Claude-powered support assistant for a fintech scale-up

Claude · GenAI on AWS

Client: a fintech scale-up offering multi-currency accounts to around 400,000 retail customers.

The challenge

Support volume was growing faster than headcount, with long first-response times outside office hours. Previous chatbot attempts had failed because answers were generic and occasionally wrong — unacceptable in a regulated financial product where incorrect guidance carries real compliance risk.

The solution

We built a retrieval-augmented assistant on Claude Sonnet 4.5 via Amazon Bedrock. Help-centre articles, policy documents and product release notes are chunked and indexed in Amazon OpenSearch Serverless with hybrid (vector plus keyword) retrieval; every answer must cite its sources, and responses touching regulated topics are routed to a human with a drafted reply rather than sent automatically. An evaluation harness of 1,200 curated question-answer pairs gates every prompt and retrieval change, and prompt caching keeps per-conversation costs low. Guardrails include PII redaction before inference and full audit logging of each response and its citations.

Claude Sonnet 4.5Amazon BedrockOpenSearch ServerlessHybrid RAGPrompt cachingAWS LambdaEvaluation harnessPII redaction
46%of inbound conversations resolved without human escalation
9h → <1minmedian first-response time outside office hours
+11ptssupport CSAT within three months of launch

Agentic document processing for a B2B SaaS company

Claude · Agents on Google Cloud

Client: a B2B SaaS company whose onboarding team manually reviewed supplier contracts and compliance documents.

The challenge

Every new customer required a review of dozens of contracts, certificates and questionnaires in inconsistent formats and several languages. The manual process took a specialist team up to two weeks per customer, created an onboarding backlog, and its findings were locked in free-text notes that could not be queried or audited.

The solution

We designed an agentic pipeline using Claude Sonnet 4.5 on Vertex AI, with Claude Haiku 4.5 handling high-volume classification and triage. Documents arrive in Cloud Storage, are OCR-processed with Document AI, then a tool-using Claude agent extracts obligations, dates and risk clauses into a validated schema, cross-checks them against the client's policy rules, and flags low-confidence items for human review in a purpose-built queue. Extraction quality is tracked continuously against a hand-labelled golden set. We also ran a Claude Code enablement programme for the client's engineers, who now maintain and extend the pipeline themselves.

Claude Sonnet 4.5Claude Haiku 4.5Vertex AIDocument AICloud RunCloud StorageTool use / agentsClaude Code
10 days → 2 daystypical customer onboarding review time
96.4%field-level extraction accuracy against the golden set
11 weeksfrom first workshop to production go-live
Technologies

Technologies we work with

We build on the platforms our clients already trust, and we keep our expertise current across all three.

Amazon Web Services Google Cloud Anthropic Claude

Product and company names are the trademarks of their respective owners and are listed here to describe the technologies we work with.

Contact

Let's talk about your next build

Tell us what you are trying to ship — a migration, a data platform or an AI product — and we will come back within two working days with a considered view on how we would approach it.

Email

We reply to every serious enquiry, typically within two working days.

Company information

Nagami Ltd is a private limited company incorporated in England and Wales.