Services

AI and LLM: advisory and build

I help teams move from “AI curious” to AI in production. That means agents that actually run business workflows, shipped and maintained, with the engineering discipline production systems need: typed outputs, cost tracking, retries, observability.

Most teams I talk to are stuck between prototype and production. The demo worked, the board is asking about AI strategy, and the path from there to something customers or staff rely on daily is unclear. That gap is exactly where I work.

The advisory and the build come from the same person. The architecture decisions I recommend are ones I've implemented myself, and if we move from roadmap to build, you keep working with me.

What I've shipped

I built a production multi-agent system end-to-end: specialised agents for conversation analysis, booking extraction, and business context, routed through a shared provider registry, with every output schema-validated. It runs Claude via AWS Bedrock with model choice matched to task complexity and cost, agent invocations as retryable background jobs, and per-user cost attribution.

Before that, I built LLM agents that automated a full agile cycle, sprint planning, backlog refinement, daily standups, retrospectives, and release documentation, plus RAG tooling for internal knowledge retrieval. Combined, those drove a productivity lift of around 25% across a 20+ engineer team.

I've also built RAG systems grounded in live data, using OpenAI function calling with anti-hallucination constraints and batch ingestion pipelines. Multi-provider is the default: OpenAI, Anthropic, AWS Bedrock. I can tell you where each actually shines, from production experience rather than benchmarks.

Engagement shapes

• AI strategy workshop (1 to 3 days): figure out what's worth building, and what isn't.
• Roadmap and architecture (2 to 4 weeks): scope the work, design the system, de-risk the build.
• Agent or RAG proof of concept (4 to 8 weeks): something real to learn from, built to graduate into production.
• Production build (6 to 12 weeks): take it live, with the operational rigour above.
• Ongoing AI advisor (a few hours a week): a senior AI voice inside your team.

Recent work

• LLM agents in production: a multi-channel payments platform
• A RAG product recommendation engine grounded in live data

Start with a conversation

Thirty minutes on where you are with AI and where you want it to be. If I can help, I'll propose a shape. If the honest answer is that you don't need this yet, I'll say so.