// service — ai engineering
AI agents & the
data that grounds them _
Most AI projects fail on the data, not the model. I build the full stack — custom Claude agents, skills and MCP servers — and the RAG datasets that make them accurate, current and trustworthy.
▸ agent.init(skills=4) loading mcp tools...... ok retrieving context..... 12 docs grounding response..... ok ▸ task.complete ✓ 0 hallucinations
// the build
From prompt to production
Custom Claude Agents
Purpose-built agents that do real work — research, triage, content ops, data entry. Full cycle: design, tools, prompts, guardrails and deployment.
Skills & MCP Servers
Reusable Claude skills and Model Context Protocol servers that connect agents to your APIs, databases and internal tools — securely and predictably.
Workflow Automation
End-to-end automations that replace repetitive manual flows. Humans stay in the loop where it matters; the boring parts run themselves.
RAG Dataset Engineering
The unglamorous work that makes AI reliable: collecting, cleaning, chunking and embedding domain data into retrieval sets your agents can trust.
Knowledge Pipelines
Ingestion pipelines that keep your vector store fresh — scraping, parsing, deduping and re-indexing on a schedule, with quality checks built in.
Evaluation & Guardrails
Eval harnesses, grounding checks and safety rails so you know an agent works — and keeps working — before it touches production.
// the part everyone skips
Building a RAG dataset
Retrieval-augmented generation is only as good as the corpus behind it. Here's the pipeline I run to turn raw, messy source material into a dataset an agent can actually rely on.
- 01
collect
Source domain documents, transcripts, APIs and structured records.
- 02
clean
Normalise, dedupe and strip noise so the model learns signal, not clutter.
- 03
chunk
Split content into retrieval-sized passages with metadata that preserves context.
- 04
embed
Generate embeddings and index into a vector store tuned for your queries.
- 05
ground
Wire retrieval into the agent and evaluate answers against the source of truth.
// adjacent service
AI media & citations
The same RAG thinking applies to marketing. I structure and seed content so answer engines and AI assistants surface — and cite — your brand. It's SEO for a world where the search box talks back.
// toolbox
What's under the hood
// let's automate it
Have a workflow begging to be an agent?
Tell me what eats your team's time. I'll tell you honestly whether an agent is the answer — and if it is, I'll build it end to end.