AI Models
for Scripture

Three-model pipeline for classifying theological statements as orthodox, heterodox, or denominational distinctive based on the first six ecumenical councils.

Two-model pipeline for discovering biblical cross-references and classifying connection types (direct quote, allusion, typology, prophecy fulfillment, and more) across 31,102 KJV verses.

Two-model seq2seq pipeline for morphological parsing and interlinear glossing of biblical Hebrew and Greek. Parses word-level grammar and generates word-by-word English translations from Macula data.

Seq2seq classifier for textual criticism — analyzes Greek NT manuscript variants across 6 critical editions (NA28, SBL, RP, ST, WH, Treg) to identify omissions, substitutions, and spelling differences.

Sentence embedding model for comparing how 5 English Bible translations (KJV, ASV, YLT, Darby, AKJV) render the same verse. Detects semantic similarity and translation divergences across 31,103 verses.

Named entity recognition model for identifying people, places, groups, divine beings, and artifacts in biblical text. Trained on KJV with entity spans and types.

Semantic search model for finding biblical passages by topic. Fine-tuned on Nave's Topical Bible and TSK cross-references to retrieve relevant verses for any theological query across all 66 books.

Rates biblical passage reading difficulty and generates simplified paraphrases for complex verses. Helps new readers and children access challenging Scripture passages.

Three-model RAG pipeline for answering apologetics questions with Scripture. Intent classifier routes questions through 5 categories, retriever fetches from 16K+ passages (Spurgeon, GotQuestions, church fathers, creation science, miracles), and Qwen3-14B LoRA generator (perplexity 4.08) produces grounded responses.