AI built on the structure of truth.
IAMDATA designs AI architecture that takes the creation principles of the Word of God as its backbone — so that answers about Scripture are grounded in a verified map, not in statistical guesswork.
General-purpose AI was not built for the Word.
Large language models are trained on everything, so they answer about Scripture with the confidence of a generalist. When the subject is faith, a confident error is more than a mistake.
When AI models took a seminary-level Bible exam, they stumbled on basic theology.
Reported by Kookmin Ilbo, March 27, 2026 — an experiment by Prof. Kim Soo-hwan of Chongshin University, using questions from the 2024 early-admission exam.
- Verse-level slips. On 2 Corinthians 11:24, a model's answer of “39” was marked wrong against the exam key.
- Blended doctrine. Answers on salvation mixed teachings from different religions into one fluent paragraph.
- Misread names. One model misidentified Jerubbaal, a figure in the book of Judges.
- Little benchmark coverage. Only 2 of the 2,500 questions in Humanity's Last Exam concern religion.
Give the model a map, then check its work.
We don't rewrite the user's question and we don't retrain the model. Before the model answers, we look the question up in a structured knowledge graph of Scripture and hand it verified context. After it answers, we check the result again. Step through the pipeline below.
ILLUSTRATIVE WALKTHROUGH — NOT A LIVE OUTPUT
Rules
The system prompt carries the structure: how Scripture is read, and what an answer must respect.
Facts
The verified passages and relationships relevant to this question, drawn from the graph.
Untouched
The user's own words stay the user prompt. The architecture works around the question, never inside it.
One graph, sixty-six books.
The knowledge graph is organized around the structure of the canon itself. Hover or tap a book, or choose a group, to see how the ring is laid out.
LOGOS
A Bible app whose answers are traced to the text, in Korean and English.
Phase 1 — in development- 01Korean & English Bible. Read and ask in either language.
- 02Knowledge-graph answers. Built on a graph spanning all 66 books.
- 03Double verification. Checked before the model answers and again after.
- 04Fewer hallucinations. Answers stay anchored to what the text says.
- 05Your choice of model. Works with GPT, Claude and Gemini.
Built, not just proposed.
Two working prototypes already run on infrastructure we operate ourselves, using open-source components and open scholarly data.
A private AI we host ourselves
A self-hosted assistant built on Open WebUI and Ollama with a Korean-capable open model (EXAONE), reachable securely from a phone or PC. It is the testbed for our architecture.
Hebrew text analysis
A Hebrew analysis tool running on NEXA GPT, built on open data sets such as the Open Scriptures Hebrew Bible, BDB and Strong's.
Four phases, one direction.
App + graph
LOGOS ships as an Android app with an initial graph and API access.
Deeper graph
The graph is expanded and refined with expert review.
Graph + API + SSL
Broader API and the start of Seven Spectrum Learning.
SSL at scale
SSL becomes the main line of the architecture.
Plainly answered.
Does LOGOS replace pastors or theologians?
No. It is a study aid that keeps answers anchored to the text and shows its grounding. Interpretation and pastoral care remain human work, and expert review is part of our graph-building process.
Which AI models does it work with?
The architecture sits around the model, not inside it, so it is designed to work with the major commercial models — GPT, Claude and Gemini — as well as open models we host ourselves.
Do you retrain or fine-tune a model?
No. We supply structure and verified context around an existing model, and verify the result afterwards. The user's question is never altered.
Is LOGOS available yet?
LOGOS is in Phase 1 development. Early testing builds are planned for Android first.
How can I support or partner with IAMDATA?
Write to us. We are glad to talk with ministries, partners and supporters, and share detailed materials after a conversation.
Let's talk.
For partnership, ministry or supporter inquiries, write to us. Detailed materials are shared after a conversation.
contact@iamdata.us