Description

- Designs systems that are correct under partial failure -- knows the failure modes of event-driven choreography (out-of-order delivery, duplicate events, poison pills, consumer-group rebalance) and builds for them
- Deep experience with Kafka or equivalent durable event log -- topic partitioning strategy, compacted topics, consumer-group semantics, exactly-once producers, transactional producers/consumers, tiered storage trade-offs
- Designs two-bus architectures -- a low-latency real-time bus (NATS JetStream or similar) alongside a durable ordered log, with a transactional outbox binding them
- Experience with backpressure, flow control, and throughput tuning in event-driven pipelines
- Defines and enforces canonical event contracts -- schemas, versioning (semver per event type), anti-corruption layers at partner boundaries
Event Sourcing, CQRS & Domain Patterns -- Expert
- Designs event-sourced aggregates from scratch -- the event stream as source of truth, fold-over-stream for current state, snapshot strategies, schema evolution under append-only constraints
- Deep understanding of the aggregate as write-side consistency boundary – why the command goes through the aggregate, not around it
- Designs CQRS read models (projections) that consume from the event log, with clear projection rebuild, catch-up subscription, and consistency guarantees -- and explicitly documents that read models enforce nothing
- Designs and implements saga patterns for irreversible multi-step commits (payment -> policy number -> issued fact) with idempotency keys, compensation registration, and retry semantics
- Defines optimistic concurrency strategies at the aggregate level – monotonic version per aggregate, version-check-on-append, stale/duplicate/out-of-order rejection with precise error
Operator & Reconciliation Patterns
- Designs application-level reconciliation loops -- desired state vs actual state in a queryable relational ledger, watch/timeout/retry/rollback as first-class domain code
- Understands the Kubernetes Operator pattern conceptually (CRDs, controllers, reconcile loop, leader election) and can adapt it to a non-Kubernetes application-level service
- Designs config-down / readiness-up interfaces -- how a commissioning authority pushes versioned configuration to components and reconciles their acknowledgements without probing
- Defines self-validating readiness -- a component that checks its own readiness at bind time and returns one honest signal, not a vague timeout

API, Contracts & Data Formats
- Designs AI-native REST APIs with JSON Schema (2020-12) + OpenAPI -- context-carrying, agent-consumable, with per-field semantics and provenance
- Understands JSON-LD and linked-data principles -- self-describing, machine-authorable documents (blueprints are JSON-LD); knows how @context, @type, and @id work
- Designs JSONata transform pipelines -- inbound and outbound boundary transforms as config (not compiled code), so a mapping change is a config change and never a redeploy
- Defines CloudEvents envelope standards for the platform -- envelope metadata, trace-context propagation, canonical-event typing and semver, routing conventions
Security -- Architecture Level
- Designs zero-trust service meshes -- no implicit network trust; mTLS between services, identity at every hop
- Architects hybrid authorisation -- coarse RBAC at the edge (OAuth2/OIDC), fine-grained state-scoped authorisation co-located with the aggregate (embedded policy-decision library, no per-action network hop)
- Designs URP (User-Resource-Permission) models with role x action x resource x state tables -- understands how role tables are authored, versioned, and fed to embedded PDPs rather than a central authorisation service
- Defines PII/PHI classification as a data model property -- tagging, masking, exclusion from logs and exports at the field level, not the service boundary

AI Integration -- Architecture Level
- Designs embedded per-component AI agents that author ComponentConfiguration from blueprint intent -- understands the control-time vs runtime boundary (agents author config; they are never on the transaction path)
- Designs AI-native APIs -- APIs structured so that an LLM agent can reason over the data model rather than just push values; context-carrying schemas
- Understands provider-configurable LLM backends -- how to abstract the model provider so the platform is not coupled to one vendor (OpenAI, Anthropic, Azure OpenAI, Bedrock, etc.)
- Familiar with structured output, tool-calling, and multi-step agentic patterns -- can design a human-in-the-loop review step as a first-class part of an AI-assisted pipeline

Languages & Polyglot Architecture
- Proficient in two or more of: Java (17+), Kotlin, Go, TypeScript/Node.js, Python
- Makes and defends language selection decisions per component -- knows when Go's binary size and concurrency model suits the runtime shell, when JVM's ecosystem suits a domain-heavy engine, when Python suits an AI pipeline
- Defines the shared shell as a cross-language pattern -- what must be identical in every language runtime, what can be language-idiomatic, and how to test uniformity
Infrastructure & Platform Engineering
- Designs Kubernetes-native deployments -- custom controllers or application- level operators, stateful vs stateless component topology, HPA for data workers, singleton control workers
- Defines cloud-agnostic infrastructure behind ports -- event broker, relational DB, object store, KMS, secret store accessed through interfaces so a cell can re-target cloud providers without re-architecting
- Writes Infrastructure as Code (Terraform, Helm, Kustomize) as a first-class engineering artefact -- cell-template onboarding, IaC conformance gates in CI - Designs two-lane observability -- OpenTelemetry (traces/metrics/logs,
PII-scrubbed at the collector) for engineering, and a business/regulator lane (projections over the canonical event log) -- correlated via CloudEvents trace-context + OTel trace IDs
Testing & Quality
- Designs integration test harnesses that run against the real event backbone --not just mocked brokers
- Defines contract testing at canonical event boundaries (schema-based or Pact) as a CI gate
- Defines AI-generated end-to-end scenario suites as a product publish gate -- scenarios are a delivery, not a report
- Writes and requires mutation testing or property-based testing for domain cores (rating, rules, premium maths) where correctness is non-trivial

Key Skills

Education

Bachelor's degree