LangGraph development,
running in production
We build multi-agent systems on LangGraph and run one ourselves: Kai, the assistant inside KAMPALO, is a LangGraph supervisor routing to specialist agents for search, analytics, Google Ads and Meta. The same patterns — typed state, checkpoints, interrupts for human approval, and evaluation on every change — go into client builds.
LangGraph work we do
Production concerns we handle
KAMPALO
Supervisor + specialist agents on LangGraph over synced Google Ads, Meta, GA4 and Search Console data. Humans approve anything that spends.
Case study →AuditQA
Tools-first financial statement review with an LLM cross-check and a learning loop from reviewer feedback.
Case study →Coppice
Codebase governance for Cursor: an architecture contract that keeps the next coding agent honest. Runs on the developer’s machine.
Case study →Common questions
What is LangGraph used for?
LangGraph is a framework for building stateful, multi-step AI agents as graphs. It is used when an agent needs branching logic, several specialist agents, memory across steps, or a pause for human approval before it continues.
Do you have LangGraph running in production?
Yes. KAMPALO’s assistant, Kai, is a LangGraph supervisor with specialist agents for search, analytics, Google Ads, Meta Ads and Meta insights, exposed through MCP tools and served over HTTP, SSE and WebSocket.
Should we use LangGraph or a simpler agent loop?
For one job with a few tools, a simple tool-calling loop is often enough. LangGraph pays off when you need multiple agents, durable state, retries or human-in-the-loop steps. Discovery ends with a written recommendation either way.
Can you take over or fix an existing LangGraph project?
Yes. We start with a short review of the graph, state, tools and evaluation coverage, then propose fixes in two-week increments.