Oxcart

Costa Rican philately has a documentation problem. 162 years of stamp history, starting with the first issue in 1863, sits scattered across hard-to-search literature: catalogs, decades of society bulletins, monographs, forgery studies, postal bulletins and auction results. Most of it is scanned PDFs, about 70% in Spanish, that never cross-reference each other and follow several incompatible catalog-numbering systems. Oxcart makes that corpus answerable.

Philatelic questions are hard because they mix exact nomenclature (catalog numbers, denominations, perforation gauges) with narrative: history, provenance, forgery analysis. A plain vector RAG isn't enough. So Oxcart parses the corpus with Dolphin, using LandingAI ADE as a selective high-fidelity fallback, embeds it into Weaviate, and fuses that with a Neo4j catalog knowledge graph. Exact queries resolve against precise entities, then expand into textual evidence, and every answer carries first-class citations down to document, page and chunk.

It's the same GraphRAG pattern we validated end to end in Canopy Intelligence and matured over years of Nairu retrieval work, from Elasticsearch to embeddings to RAG to GraphRAG, now pointed at a new domain: cultural heritage. We present it honestly: an advanced research prototype with a Gradio demo we run on demand, not a production service. MIT-licensed and open source.