Query Assist

The Problem

MQL (Matrix Query Language) is the scripting language behind ENOVIA/ 3DEXPERIENCE’s platform. Good material for it is scattered — some real example scripts here, a note there, tribal knowledge nobody’s ever centralized. Every developer ends up rediscovering the same syntax answers for themselves instead of having one place to ask.

What it does

Lets a developer just ask the question in plain language — “how do I expand a relationship?”, “what’s the syntax for a temp query?” — and get an answer grounded in a curated set of example scripts and reference notes, citing exactly which source it pulled from instead of guessing.

How it is built

A hybrid retrieval index (vector embeddings + BM25) sits over a curated set of real .mql example scripts and reference notes. The model (Gemini) gets tools to search that material and decides which to call itself, rather than always stuffing everything into the prompt. A small Streamlit UI wraps it with a password gate, source citations, and query history.

A simple tech stack workflow to understand

Example scripts + reference notes
   → chunked & indexed (vector + BM25), one-time ingest
   → question asked in the Streamlit app
   → Gemini picks a tool, retrieves the matching chunks
   → grounded, cited answer

Everything the app does after that — the answer, the citation, the usage stats, the query history — comes from that same loop: ask, retrieve, answer, cite.

Where it stands

An early-stage prototype developed to illustrate the vision for a larger initiative.