Inspiration
Faith comes by hearing the Word of God, but when people face real problems, they often do not know what biblical truth they need to hear or what unbelief is preventing them from believing God's Word.
A preacher or Bible teacher does more than provide a random verse. They identify what needs to be taught, bring together relevant Scripture and teaching, reinforce truth, and progressively help faith grow.
BibleOmnix was built to bring this kind of Scripture-based teaching journey into an autonomous AI agent.
For this hackathon demonstration, BibleOmnix focuses on Christian marriage and relationships.
What it does
BibleOmnix is a Scripture-based AI Faith Creator.
A user simply describes what they are facing, for example:
"I don't feel loved by my husband anymore."
BibleOmnix autonomously identifies the specific area of unbelief or spiritual need, determines the biblical truth that needs to be established, checks what the person has already been taught, and plans what should happen next.
The planner can decide to establish a new truth, reinforce an existing truth, deepen previous teaching, or create and switch between different teaching threads.
This means BibleOmnix does not treat every message as a new question. It remembers the person's teaching journey and adapts what it teaches next.
How we built it
BibleOmnix uses an agentic architecture built with Google ADK, Gemini 3.5, and Vertex AI.
The autonomous teaching loop is:
Discern → Plan → Retrieve → Teach → Remember → Adapt
Gemini helps understand the user's situation and teaching need. The planner determines the next teaching action based on the user's current state and previous teaching.
BibleOmnix then uses its v22 vector-based teaching engine.
The vector engine works like a library of approved Scripture-based teaching. Instead of allowing the AI to freely invent doctrine, the vector system searches the approved teaching material by meaning and retrieves the teaching components most relevant to the truth the agent has decided to establish.
In simple terms:
The agent decides what needs to be taught, while the vector engine finds the right approved teaching to teach it.
This gives BibleOmnix both autonomy and authenticity.
The selected teaching is presented conversationally, and the user's teaching state is then updated for future turns.
Autonomous multi-turn teaching
Suppose the user first says:
"I don't feel loved by my husband anymore."
BibleOmnix can establish the biblical truth needed for that situation.
If the user later says:
"I understand, but what if he still doesn't change?"
the agent recognizes that this belongs to the same teaching thread. It can autonomously decide to reinforce or deepen the existing truth rather than starting over.
If the user then says:
"I also cannot forgive him for what he did."
BibleOmnix can identify a different spiritual need and create another teaching thread around forgiveness while retaining the previous teaching journey.
Our Agent Activity interface makes these autonomous decisions visible, including the identified need, teaching target, planner action, teaching-thread routing, vector teaching components, and memory update.
Challenges we faced
One of the biggest challenges was balancing AI autonomy with doctrinal authenticity.
We wanted the agent to genuinely reason about what a person needs and autonomously plan the teaching journey, without turning the system into a general LLM that could freely generate or alter the underlying teaching.
Our solution was to separate agentic decision-making from teaching-source authority.
The AI determines what needs to happen next, while BibleOmnix's vector-based system grounds the actual teaching in approved Scripture-based material.
Another challenge was multi-turn memory. BibleOmnix needed to recognize whether a user was continuing an existing spiritual issue or introducing a new one. We therefore created separate teaching threads and state tracking so the agent can progressively establish, reinforce, and deepen truths over time.
What we learned
We learned that agentic AI can be especially powerful when it does not merely generate answers, but manages a continuing journey toward a goal.
For BibleOmnix, that goal is not simply completing a conversation. The goal is progressively establishing biblical truth and building faith.
What's next
Marriage and relationships are the focused use case for this hackathon demonstration, but the architecture is designed to expand into other areas of life.
The long-term vision is for BibleOmnix to remember what a person has been taught across many areas, identify where biblical truth still needs to be established, and progressively build faith through God's Word.
BibleOmnix does not simply answer Bible questions. It determines what biblical truth a person needs next and progressively teaches toward faith.
Log in or sign up for Devpost to join the conversation.