Workflow & agent design
Graph-based orchestration with LangGraph: sequential and parallel steps, conditional routing, checkpoints and retries.
AI Agents & Automation
Agents that do real work across your tools, safely.
I design and build AI agents that take on multi-step work across your systems: reading and writing to your tools, calling APIs, and asking a human before anything risky. Built on LangGraph and the Model Context Protocol, with guardrails, audit trails and cost controls from day one.
Sound familiar?
Problems this solves
Repetitive, multi-step work across several systems is eating your team's week.
Your agent prototype impressed in a demo but breaks on real inputs.
You can't see what your agents did, why they did it, or what they cost.
What you get
Scope of work
A typical engagement covers the following, tailored to your stack and goals.
Tool integrations via MCP
Model Context Protocol servers that give agents safe, standard access to your APIs, databases and SaaS tools.
Guardrails & human-in-the-loop
Input and output checks, PII redaction, and approval steps before agents take sensitive actions.
Observability & cost control
Tracing, evaluation and per-agent token metering, so you can see quality and spend per workflow.
Production deployment
FastAPI services in Docker on your cloud (Azure or AWS), including Kubernetes where needed.
Voice & chat channels
Optional speech-to-text and text-to-speech, plus delivery in Microsoft Teams, web or your product.
Use cases
Where this fits
- Triage support tickets, enrich them with account data and draft replies for review
- Internal copilots that take actions in Microsoft Teams or Slack
- Research agents that gather sources and produce structured reports
- Back-office automation across CRM, ERP and spreadsheets
Toolkit
Technology I use
- LangGraph
- LangChain
- MCP
- OpenAI Agents SDK
- FastAPI
- Azure OpenAI
- Anthropic Claude
- AWS Bedrock
- LangSmith
- Langfuse
I pick tools to fit your environment and constraints, not the other way round. Models and vendors stay swappable.
Related experience
Built this before
From enterprise platforms I've built. Client names are withheld.
FAQ
Questions about ai agents & automation
How do you stop an agent from doing something it shouldn't?
Agents only get the tools they need, with scoped permissions. Sensitive actions go through an approval step, and every input, tool call and output is checked by guardrails and logged so you can audit it.
Which models do you use?
Whatever fits the job and your constraints: Azure OpenAI, OpenAI, Anthropic Claude, Google Gemini, AWS Bedrock or open-source models. The architecture keeps the model swappable.
Talk to me about ai agents & automation
Tell me what you're trying to build. In a free 30-minute call we'll work out whether AI is the right tool, what it would take, and whether I'm the right person to build it.