About the Egyptian Inscriptions Database

How to read a record

This database aggregates and enriches other projects' data, so it is explicit about where every fact comes from. On each object page you will see, clearly distinguished:

• From the source institution (authoritative): accession, description, dimensions, dating, and a link back to the original catalogue record. Photographs we have archived carry a caption naming the holding institution and the licence.
• AI-inferred (machine-generated, may be wrong): the image-analysis panel reading deities, royal names, materials, and signs from the object's photograph.
• Approximate location: most map points sit at the site centroid, not the exact findspot, and are labelled as such.
• Inferred links: cross-references we matched ourselves (rather than ones stated by the source) carry their match method and a confidence tier.

The goal is trust-but-verify: take what is useful and stay one click from the canonical source. Found an error? Use the Feedback button on any page.

Sources and modalities

The corpus draws on three distinct kinds of source:

• Text corpora — ORAEC for hieroglyphic texts with Manuel-de-Codage transliteration and scholarly German translations; the Duke Databank (DDbDP), HGV and APIS for documentary papyri; Trismegistos for the cross-source spine of text and place IDs.
• Museum catalogues — the Metropolitan Museum, British Museum, Brooklyn Museum, Berlin, Cleveland, Walters, Art Institute of Chicago, Fitzwilliam, ISAC Chicago, Freer (NMAA), Manchester, and Leiden. Their records carry accession numbers, curator descriptions, sometimes Wikidata Q-IDs, and (when present) inscription transcriptions.
• Excavation archives — the Egypt Exploration Society's Tell el-Amarna corpus (Petrie-grid-anchored objects) and the Harvard/MFA Giza Archives.

Inscriptions land in any of three content modalities, side-by-side on the same row: hieroglyphic transcription (MdC; sometimes Greek/Coptic/Arabic for non-pharaonic text), English-prose curator description, or structured sign-name vocabulary as JSON.

The cross-source spine

Many objects carry cross-references to other sources where the same physical artifact or text appears. Trismegistos-Text IDs are primary for papyri — a single TM-Text ID stitches together a DDbDP edition, its HGV metadata record, and an APIS collection record of the physical papyrus, including photographs from the Duke Papyrus Archive when those exist. Some IDs also reach a museum photograph (Berlin inventory matches and Wikidata P8532). Other museum objects link by Wikidata Q-IDs and accession numbers. The corpus today has 675,233 cross-references across all link types.

AI analysis

Objects with directly-fetchable JPEG images go through a structured vision-AI pass (currently GPT-5.4-nano, grounded on the museum catalogue record). The model identifies deities, royal cartouches, materials, visible hieroglyphic signs (with Gardiner codes when applicable), and transcribes any plainly-visible text. 24,801 images have been analyzed so far. Output is stored as structured JSON and surfaces on each object's detail page in the AI panel; the model name is printed on that panel.

Inscriptions that have a transcription but no scholarly English translation are also passed through an AI translator. AI-generated translations are flagged in blue on the object page; scholarly translations from the source materials are flagged in ochre. Most Egyptian-language translations in the corpus are scholarly German (from ORAEC and TLA), not English.

Schema

The canonical schema is documented in docs/schema_v1.md. Postgres with the pgvector and PostGIS extensions; 1,536-dim OpenAI embeddings on every inscription for cross-modality semantic search; EPSG-4326 points on objects and excavation contexts for the map view.

Reuse and citation

Every object has a stable URL: /object/{id}. Inscription and translation content carries the source's license at the source-table level — see the Sources page for per-source licensing. The database itself is a research aggregation; cite the individual source for any specific record, and this site as the aggregator if useful.

Source code on GitHub →