Claude Code, coding agents, LLM applications, RAG, APIs and workflow automation only become valuable when the surrounding engineering system is reliable. That layer — context, tools, tests, evals, controls, observability and production delivery — is where my strongest fit sits.
I use coding agents as part of a production delivery system: understand the repo, frame the task, plan the change, implement it, run tests, inspect the diff, evaluate the result, document the decision and release through controlled automation.
The same pattern extends into agentic applications: models get bounded tools, retrieval gets measured, APIs remain deterministic, and security, approvals, observability and fallback paths are designed before scale.
Hover or focus a branch for my evidence. The visible keywords mirror the role specification.
GitHub Copilot · Claude Code · code generation · intelligent SDLC
Daily Claude Code + Codex/CLI, 60+ reusable skills, MCP, subagents, tests and release automation. Copilot-specific depth is transferable rather than overstated.
OpenAI GPT · Azure OpenAI · Anthropic Claude · prompt engineering
Production Claude, OpenAI and Azure OpenAI work with routing, retries, fallbacks, prompt optimisation and cost/latency controls.
Agents · copilots · workflows · LangGraph · CrewAI · AutoGen · Semantic Kernel
AIOS, AIBO and ABC: planner/router/executor patterns, MCP tools, memory/state, typed hand-offs, approvals and recoverable fallbacks.
Python · JavaScript/TypeScript · REST · microservices · scalable architecture
Python/FastAPI plus TypeScript/Node/Next.js, REST APIs and production application architecture across enterprise systems.
RAG · embeddings · semantic search · Pinecone · ChromaDB · Weaviate · Azure AI Search
Embeddings, hybrid search, GraphRAG, LlamaIndex, pgvector, Pinecone, Weaviate, FAISS and Chroma with provenance and permission-aware retrieval.
Evaluation · observability · monitoring · AI governance · security · responsible AI
Tests, evals, tracing, auditability, GDPR/redaction, approval boundaries, provider fallbacks and human-in-the-loop controls.
Azure/AWS/GCP · GitHub Actions · Jenkins · Docker · Kubernetes · Terraform
AWS, GCP and Azure exposure with GitHub Actions, Jenkins, Docker, NGINX and CI/CD. Kubernetes/Terraform depth should be validated rather than inflated.
Enterprise integration · stakeholders · mentor developers · accelerate adoption
15+ years across Allianz, RWS, Sainsbury's and D&B/Cogniflare bridging architecture, engineering, governance, stakeholders and delivery.
Build AI-powered software with Claude Code, GitHub Copilot and modern LLMs — safely, measurably and at enterprise scale.
The useful question is not whether a keyword appears on a CV. It is whether there is evidence behind it and a clear way to apply that evidence to the job.
Claude Code is part of my daily engineering model: repo-aware analysis, planning, implementation, refactoring, testing, documentation and release workflows, supported by 60+ reusable skills, MCP integrations, subagents and hooks.
Turn coding agents into a repeatable engineering system with explicit context, review, testing, permissions and release controls rather than isolated prompts.
My deepest day-to-day coding-agent experience is Claude Code and Codex/CLI. I have also delivered Microsoft Copilot and Copilot Studio solutions in enterprise environments, so the workflow and governance patterns transfer directly.
Apply the same controlled AI-assisted delivery loop across GitHub Copilot while being precise about tool-specific experience rather than inflating it.
Production work with Anthropic Claude, OpenAI, Azure OpenAI and multi-provider model routing across live agentic platforms, including fallbacks, retries, cost/latency trade-offs and operational controls.
Choose the right model boundary per workflow and keep providers replaceable behind tested service contracts.
AIOS, AIBO and ABC use multi-agent orchestration, MCP/tool contracts, router/planner/executor patterns, memory/state, typed hand-offs, human approvals and recoverable fallbacks across a 115-role workspace.
Build bounded agents that can act through real APIs and business workflows while preserving auditability and human control where risk requires it.
Hands-on retrieval work spanning embeddings, hybrid search, GraphRAG, LlamaIndex, pgvector, Pinecone, Weaviate, FAISS, Chroma and MongoDB-backed knowledge systems.
Treat retrieval quality as an engineering problem: ingest, chunk, embed, retrieve, rerank, permission-filter, cite and evaluate against real user tasks.
Hands-on backend and product engineering with Python, FastAPI, REST/JSON APIs, TypeScript, Node.js, Next.js and React, connecting models to production services and enterprise workflows.
Ship thin end-to-end slices through the real stack so AI capability is tested as part of the application, not in a notebook beside it.
Current production patterns include automated tests, evaluation, model routing, tracing, retries, caching, provider fallbacks, audit trails, incident-aware controls, GDPR/redaction and human-in-the-loop review.
Measure task success, groundedness, regressions, latency and cost, then expose failure clearly enough that engineering teams can operate the system with confidence.
15+ years delivering production software across Allianz, RWS, Sainsbury's, Dun & Bradstreet/Cogniflare and other enterprise environments, bridging architecture, engineering, governance and stakeholder delivery.
Accelerate the team without creating dependency: establish patterns, pair with engineers, document decisions and leave reusable components and guardrails behind.
A live multi-agent platform with 115 defined roles, per-agent models/tools/memory, permissions, evaluations and provider fallbacks. The useful proof is not the number of agents; it is the operating model around them.
An engineering cockpit centred on 60+ reusable Claude Code skills covering build, test, deploy, analyse, ingest, sync and publish workflows — turning coding agents into repeatable delivery infrastructure.
Enterprise AI delivery spanning Copilot Studio assistants, voice/contact-centre automation, SharePoint/OneDrive grounding, Azure OpenAI and governance requirements including GDPR, redaction and auditability.
Agent behaviour can be flexible. Identity, permissions, schemas, API contracts, tests, deployment, logging, cost ceilings and rollback paths should not be.
Repo context, bounded tasks, tests, diff review, CI/CD and reusable agent skills rather than one-off code generation.
Retrieval quality, permissions, provenance, citations and evaluation before prompt theatre.
Typed tools, explicit state, retries, approval gates and recoverable execution paths.
Security, auditability, monitoring, failure evidence, cost/latency budgets and an owned runbook.
The fastest route to trust is one real workflow with measurable quality and a team that understands how to operate it.
Identify the highest-value developer and application workflows, current Claude Code/Copilot usage, codebase boundaries, model providers, data access and measurable acceptance criteria.
Take one real use case from repo context → agent/tool action → code/API change → automated tests → review → deployment, with observability and approvals included.
Add evaluation, regression tests, groundedness checks, permissions, secrets boundaries, audit evidence, failure recovery and cost/latency measures around the workflow.
Turn the first successful loop into reusable skills, prompts, tools, templates, CI checks and team guidance so adoption compounds instead of fragmenting.
Is the primary outcome developer productivity, customer-facing AI applications, or both?
How far has the Claude Code / GitHub Copilot rollout progressed today — individual usage, team standards, or platform-level enablement?
Which use cases would make the first 90 days an obvious success?
What are the main application stacks, cloud platforms and model providers already approved?
Where must human approval remain mandatory, and where is bounded agent autonomy acceptable?
How are AI-generated code quality, security, test coverage and production incidents measured today?
I bring the architecture, hands-on coding, agents, retrieval, APIs, evals, controls and enterprise delivery needed to turn Claude Code, Copilot and modern LLMs into a reliable production capability the wider engineering team can own.
The job-spec terms plus the production concepts needed to understand them end-to-end. Hover or keyboard-focus any tile for its explanation.