Projects
Tools for AI you can inspect.
Side projects, written on nights and weekends with Claude, and listed in the order a project needs them. LoopCloser, RAGBuilder and IBO are under the MIT licence. ABO is coming soon.
01Why these tools
After putting a dozen GenAI assistants into production, I kept running into the same gaps, and none of them was about the model. Expert knowledge stays in spreadsheets, with no shared criteria and no traceability. Knowledge bases can't say where an answer comes from, or from which version. Agents are judged only on their final answer, which doesn't show whether classification, retrieval or drafting went wrong. Each tool closes one of these gaps, and in regulated sectors, closing them is how trust becomes measurable.
02Phases
LoopCloser
CurationRaw data, curation and expert knowledge extraction.
Where a project starts. Domain experts review and label raw data against configurable criteria, so what they know becomes a curated dataset.
RAGBuilder
RAG databasesBuild, audit and industrialise the deployment of RAG databases.
Versioned, inspectable RAG tables in PostgreSQL + pgvector. Edit, evaluate and ship retrieval knowledge as auditable data products, not black-box indexes.
IBO
AgentsBuild, audit and industrialise agent deployment.
Intelligent Back Office, a test bench for LangGraph agents. Experts watch agents run node by node and give feedback on every step, and teams re-run from any node to test a change in seconds.
ABO
Operate and scaleComing soon.