What we build

Fourteen capabilities, each proven by software that is live

Every capability below names the product that proves it. Where a system is built but not yet in production, the badge says so.

AI

AI product development

Agents, agentic systems, retrieval and the pipelines underneath them — engineered with tests, explicit tool grants and a human gate on anything irreversible.

AI agent development

A defined role with tool access and explicit permissions, built for bounded, repeatable judgement work inside a real workflow.

Agentic AI systems

Multi-step workflows and multi-agent systems on CrewAI, LangGraph and LangChain: routing, validation and orchestration held in deterministic code.

AI-powered SaaS

Products where AI does the data entry, the reading and the noticing, and a person makes every decision that matters.

Retrieval and RAG

Chunking, embedding, hybrid vector and keyword search, reranking and metadata filtering, with a relevance threshold that decides when the model should not be called at all.

Document and data pipelines

Ingestion, OCR and vision extraction, validation, typed persistence and indexing. A field the model is not sure of comes back empty and flagged, never plausible.

Engineering

SaaS and software engineering

The platforms AI has to live inside: identity, permissions, billing, audit, integrations and the mobile and web surfaces people actually touch.

SaaS product engineering

Multi-tenant platforms where every customer gets their own branding, catalogue, rules and domain as configuration, not a fork.

Custom software

Systems of record for work that cannot be switched off: collection, reconciliation, ledgers and the reporting a regulator or an auditor will ask for.

API and integrations

Money transfer, AEPS, recharge, bill payment and ticketing on one login — several providers behind one contract, with ledger reporting.

Mobile applications

Native and cross-platform applications shipped to the App Store and Google Play, over a shared backend contract. Flutter and React Native, both in production.

Web applications

Web applications, APIs, backend services and content platforms — including work with no AI in it at all, where that is the right answer.

What a multi-tenant platform has to get right, and what we design in from the first sprint:

Tenancy modelIdentity and SSOPermissionsData isolationBilling and entitlementAudit trailsWhite-label surface
Infrastructure

Cloud and DevOps

A system that cannot be deployed safely, observed honestly and paid for predictably is not finished, however good the model is.

Cloud & DevOps

Architecture, migration, delivery pipelines, containers, observability and cost control on 4 cloud platforms. Your accounts, in your name, from day one.

Modernisation

The system keeps running while you modernise it

The proposal you usually get is a rewrite. It is the wrong answer for a platform that is currently processing transactions, because the risk sits entirely with you and the value arrives only at the end.

We attach at a seam instead. New capability reads through the same API contract your existing clients use and acts only through explicit grants. Switch the layer off and the platform runs exactly as it did before.

Reversible by design. The agent layer attaches at one seam. Nothing is threaded through the application, so removing it is a configuration change rather than a project.
Strategy

MVP, modernisation and automation

Where to start, how to add AI to something that already works, and what to automate before writing a line of application code.

MVP development

A first working version built to test the idea against real users and real data, on a foundation that does not have to be thrown away. No-code where that is the honest answer.

Enterprise and modernisation

The platform keeps running while AI attaches at one seam: it reads through the existing API contract, proposes rather than executes, and switches off as a configuration change.

AI automation and workflows

Seven chained agents, each with a typed output schema, behind an approval gate that pauses the run and resumes on a person’s decision.

The stack

The layers we build in

Chosen per workload rather than per habit. The architecture is designed so that every one of these decisions stays reversible.

LayerTechnologies
BackendPython · FastAPI · Node.js · PHP · Java · .NET
DatabasesPostgreSQL · SQLAlchemy · asyncpg · Alembic
Async & schedulingCelery · Redis · Flower · APScheduler
MobileFlutter · React Native · Kotlin · iOS
FrontendReact · TypeScript · Vite · Next.js · Tailwind
Cloud platformsAWS · Microsoft Azure · Google Cloud · DigitalOcean
Cloud engineeringNetworking & VPC design · managed databases · object storage & CDN · secrets management · autoscaling · cost control
AI orchestrationCrewAI · LangGraph · LangChain · crew-manager pattern
Agent toolingMCP (Model Context Protocol) · function and tool calling · typed task contracts · permissioned tool grants
LLM providersOpenAI · Anthropic · Google Gemini, behind one adapter
Vector & retrievalQdrant in production. pgvector, FAISS and managed stores where the workload suits them. Hybrid search, reranking, metadata filtering, chunking strategy
CI/CDGitHub Actions · GitLab CI · Bitbucket Pipelines · build, test and deploy on every merge · staged environments · rollback
Containers & runtimeDocker · Docker Compose · Kubernetes · Nginx · managed container services
Infrastructure & observabilityInfrastructure as code · Terraform · centralised logging · metrics · alerting · uptime and cost monitoring
Qualitypytest · vitest · held-set evaluation · architecture decision records
SecurityAccess control · audit trails · ISO 27001 practice

Tell us what has to be built

Sixty minutes on the requirement, the stack and the shortest honest path to something real in front of users.

A 60-minute architecture review · no charge · the notes are yours either way