Audit before you ship
Independent review of your retrieval / KG / agent stack. We bring our taxonomy and our own tools and tell you, in writing, where it breaks.
Consultancy · tools · open critique
AI plumbing that holds up under inspection.
We help teams ship retrieval, knowledge-graph, and agent systems that survive contact with real evaluation. Two open products, one weekly research feed, and a public method page.
What we do
Three things, deeply
We pick problems where evaluation is the hard part and the literature is loud. Then we build the tooling and write the critique while we work.
Build the missing layer
Concept grounding, schema constraints, ontology validation, evaluation harnesses. The unsexy plumbing that keeps the rest honest.
Stay current with us
We track the field weekly across arXiv and adjacent feeds, publish a per-paper first read, and collect failure patterns. You get the synthesis without the doomscroll.
Properties
Three public surfaces
Each of these is a working tool you can use today and a portfolio piece showing how we think.
Curated feed of papers in retrieval, KGs, ontology, and agent safety. Each entry has a short machine-written first read, citations, and a weekly thematic digest.
Open the tracker →Failure catalog for weak GraphRAG and graph-AI claims. Auto-published critiques pass a public quality gate; every entry shows its source quotes and is retractable.
Open GraphSlop →Concept grounding and disambiguation against SUMO and custom ontologies. Useful when you need extracted entities to mean the same thing across runs.
Open toslop →How we publish
Machine-first, retractable, in the open.
A local model writes the first read. A schema validator and a quoted-line verifier check it before anything ships. Every machine-written entry carries an auto badge and a "Report this" link. Mistakes get a public retraction with a strikethrough and a note.
That means our research feed updates as fast as papers land, and our credibility is the audit trail, not the volume.
Read the method →