Who Trained Rafeeqy? — Even if you say "we just use open-source models," you have built nothing.
When the evidence shows that "open source" does not describe Rafeeqy — no GitHub repository, no published model, no recognized license — the most likely next move is a retreat: "We never claimed we built the model from scratch. We meant we use a publicly-available open-source model — like LLaMA or Mistral or Qwen."
This page is written for that retreat specifically. Because, if it comes, it opens a deeper problem than the first: if that is in fact what you do, then you are not the developer of a Christian AI. You are a user of a generic AI, wrapping it in a Christian skin.
1. You did not train the model
If Rafeeqy is using LLaMA from Meta, or Mistral from Mistral AI, or Qwen from Alibaba, or any other open-source model published by a team other than yours — then you did not choose its training data. You did not theologically review it. You do not know what that model read about Christianity, about other religions, about every heresy, every philosophical current, every Bible-related debate that has ever appeared on the internet.
The model arrived to you fully loaded with everything it had learned. You were not in the training room. You received it ready-made. What you know about its data is what anyone in the world knows by reading the public model card. No more, no less.
2. What you actually do = what every GPT-4 or Gemini developer does
Assume Rafeeqy does what any modern AI application does:
- System Prompt: A piece of text you write to steer the model toward a Christian voice. This is standard practice for every developer building on GPT-4, Gemini, or Claude. There is nothing technically Christian about it.
- RAG (Retrieval-Augmented Generation): A pipeline that retrieves Bible passages or theological references and injects them into the prompt alongside the user's question. This technique has been mainstream since 2023 and is used by thousands of applications across every industry. There is nothing technically Christian about it either.
- Fine-Tuning: If you did fine-tune on Christian texts, fine-tuning adjusts a model's surface behavior. It does not rebuild what the model knows underneath. The base remains.
These three techniques — system prompts, retrieval, and fine-tuning — are what every developer building on top of an AI model uses, whether the model is closed (GPT-4, Gemini, Claude) or open (LLaMA, Mistral). There is no engineering difference.
Rafeeqy = an open-source model (which you did not build) + a system prompt (which any developer writes) + RAG over the Bible (a generic technique) + a marketing skin.
A Christian ChatGPT app built by any solo developer on GPT-4 = a closed model + a system prompt + RAG over the Bible + a skin.
The engineering difference: which base model is used. The moral difference: one of these does not claim to be "the first Christian AI." The other does.
3. So what is "Christian" about it?
- Not the model — you did not build it.
- Not the training data — you did not select it.
- Not the architecture — it is the same Transformer architecture as every modern model.
- The only "Christian" thing here is: the text of the system prompt, the sources fed into RAG, and the words of the marketing.
This is exactly what hundreds of applications around the world do: a wrapper on a generic model, scoped to a subject. Some target coffee lovers; some target lawyers; some target doctors; some target believers. None of them claim to be "the first AI" in their domain. Because the claim would be false — technically.
4. What a real "Christian AI" would actually require
If you wanted to build a Christian AI that genuinely earns the name, you would need:
- To train (or substantially re-train) a model on a theologically-reviewed corpus — Scripture, the Fathers, the councils, recognized commentaries. This is expensive (significant GPU hours, time, research expertise).
- To disclose the data — which texts specifically, from which Christian tradition, reviewed by whom.
- Independent theological evaluation — a named committee of qualified reviewers.
- Disclosure of the base model and version — so anyone with technical literacy can verify.
- An evaluation methodology — tests against theological questions, with published results.
- An error-correction mechanism — when a user discovers a theological mistake, how it is reported and remedied.
None of this is on the Rafeeqy site. Not the base model. Not its data. Not the reviewers' names. Not a methodology. Not an error-correction process.
5. Bottom line
You are not the developer of a Christian AI. You are a user of a generic AI model, wrapping it in a Christian skin with a System Prompt and RAG. This is what any startup using GPT-4 for its own niche does.
The honest description: "A Christian-themed application, built on top of the [model name] model developed by [company name]."
The description you use: "The world's first open-source Arabic Christian AI."
Between those two descriptions: the distance of the truth.
This investigation does not deny your right to build a Christian-themed application on a generic model. That is a legitimate, common, useful thing to do. What it denies is your right to call that "building a Christian AI." Because what you have built is a skin — not an intelligence.
"Prove all things; hold fast that which is good."