GADIEL MUÑOZ
← Back to work
● In development

AskDocs

Semantic search and RAG chat running 100% inside Oracle AI Database 26ai. Every answer cites the exact passage of the document it comes from.

Role
Design, data and development
Year
2026
Platform
Oracle APEX 26.1
Code
Private

Problem

Companies have manuals, contracts and procedures nobody can find, and they don't want to send their documents to outside services just to search them.

Solution

Extraction, chunking, embeddings and vector search all happen inside the database. Each answer links the exact passage that backs it, and the language model is optional.

Key decision

The suggested embedding model passed every technical check, but it was trained on English. I caught it by measuring search ranking on real Spanish documents, and switched to a multilingual model.

English model

3/6

correct searches in Spanish

Multilingual model

6/6

correct searches in Spanish

Stack

Oracle AI Database 26aiAPEX 26.1AI Vector SearchONNX · multilingual-e5-smallPL/SQL

Backed by

Badge Oracle AI Vector Search Certified Professional
Badge Oracle AI Autonomous Database 2025 Certified Professional

Certified in Oracle AI Vector Search and Autonomous Database, the two technologies this project runs on.

Next projectVirureko