Embeddings Bíblicas PT‑BR
The first semantic embedding model for biblical search in Brazilian Portuguese
Live
The first semantic embedding model for biblical search in Brazilian Portuguese
LiveEmbeddings Bíblicas PT-BR is the first semantic embedding model for biblical search in Brazilian Portuguese. Development started in early 2026 as part of my research into AI and Christian spirituality.
Built by another similarity model on ~9,000 curated query-verse pairs, the model understands the emotional and spiritual intent behind what someone is feeling — not just keywords. When someone types "I feel like God abandoned me", it finds Isaiah 49:14. When they type "I can't stop sinning", it surfaces Romans 7:24-25.
The model runs on a custom inference stack on an Orange Pi RK3588, using the NPU via RKNN for low-latency embedding generation. A public API delivers full results including the main verse, secondary references, and a theological justification — all in a single call.
This is more than a search model. It's the foundation of a new kind of spiritual companion that meets people where they are, in their own words.
From a feeling to Scripture — in milliseconds
The user types something natural — a feeling, a prayer, a struggle — in Brazilian Portuguese. No need for keywords or Bible jargon. The model was trained to understand human emotion, not just text.
The input is converted into a high-dimensional vector that captures its spiritual and emotional meaning. This runs on-device via the RK3588 NPU using RKNN, delivering results with very low latency.
The query embedding is compared against all indexed verse embeddings using cosine similarity. The closest matches are ranked by how well they address the intent behind the input.
The public API returns a full structured response: the primary verse, secondary references, and a theological justification — all in a single call, ready to integrate into any app or service.
Built for spiritual search, not just text matching
Understands the emotional and spiritual meaning behind a query — not just surface-level keywords. Matches feelings to Scripture with remarkable accuracy.
Trained on ~9,000 curated query-verse pairs covering a wide range of spiritual experiences — doubt, grief, guilt, hope, praise, and more.
Runs on an Orange Pi RK3588 using the NPU via RKNN for fast, efficient embedding generation entirely on local edge hardware — no cloud dependency.
A clean API delivers the primary verse, secondary references, and a theological justification in a single request — easy to integrate into any project.
Built specifically for Brazilian Portuguese from the ground up — the first model of its kind designed for the Brazilian Christian community.
Every training pair was carefully selected and curated for theological accuracy and linguistic quality, ensuring the model reflects genuine scriptural wisdom.
Edge AI meets biblical intelligence
Python
RKNN / NPU
Orange Pi RK3588
Sentence Transformers
REST API
Vector Search
One request. One feeling. The right Scripture — in milliseconds.
Need a token? Get one free at /api/docs — log in with your PLVictor account.
curl -X POST https://plvictor.com/api/embed/search \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_TOKEN" \
-d '{
"text": "sinto que Deus me abandonou",
"top_k": 3
}'{
"query": "sinto que Deus me abandonou",
"results": [
{
"reference": "Isaías 49:14",
"text": "Mas Sião diz: Já me desamparou o Senhor...",
"translation": "ARC",
"theological_justification": "O povo de Israel, no exílio, achou que Deus os tinha esquecido...",
"secondary_verses": [
{
"reference": "Hebreus 13:5",
"text": "Não te deixarei, nem te desampararei.",
"translation": "ARC"
}
]
}
]
}