Expertise
AI & agentic systems
LangGraph, LangChain and CrewAI agents, retrieval-augmented generation, evals, and AI-assisted delivery.
What problems look like here
- Agent output has to be checked before anyone relies on it.
- Answers that must come only from a known record.
- Several specialised agents working under one assistant.
What I built
Case studies
-
Agentic AI systems
Agent systems I build and run daily: an intelligence-triage agent with eval-driven loops, a self-hosted multi-agent assistant, an agentic document pipeline and a two-model agentic engineering workflow.
More work
- Evidence ledger Every claim linked to its source, with deterministic document generation.
- ComfyForge ComfyUI pipelines from text to image, video and audio. Stack: ComfyUI
- Local AI inference lab Ollama and ComfyUI on an RTX 5060 Ti. Stack: Local inference (Ollama), ComfyUI
Patterns I use
-
Eval-driven loops, so an agent's changes are measured before they are kept.
Used in Agentic AI systems.
-
A primary assistant that hands work to research, development and finance agents.
Used in Agentic AI systems.
-
Human review before anything an agent drafts is sent.
Used in Agentic AI systems.
-
One model builds while another reviews, with the roles swapped by task.
Used in Agentic AI systems.
-
Answers grounded in a compiled knowledge base, with the sources cited.
Used in Agentic AI systems.
Stack for this domain
Each links to its entry on the stack page.
Ask about this
What has he built in AI and agents?
Answers come from the content on this site.
The chat needs JavaScript. Without it, ask through the contact page.