PySemantic

Like many of us wrestling with complex pipelines and massive Databricks Delta tables, I found myself constantly fighting the same battle: writing massive 500-line SQL joins, accidentally double-counting metrics (the dreaded 1:N fan-out trap), and dealing with business logic scattered everywhere.

Enterprise semantic layers are incredibly powerful, but sometimes you don't want to deploy a massive new infrastructure tool just to get centralized metrics. Sometimes you just need a lightweight, Python-native engine.

So, I built PySemantic.

It’s an open-source semantic layer that translates high-level business metrics into mathematically safe, dialect-aware SQL.

Here is what it actually does under the hood:

• Graph-Based Routing: You define your models and many-to-one relationships in Python. It uses NetworkX to automatically find the safest, most optimal join path between any two tables in your data warehouse.

• Native Fan-Out Protection: The query planner actively detects and blocks reverse 1:N traversals and cross-fact queries. It separates dimensions into WHERE clauses and measures into HAVING clauses automatically.

• Dialect Agnostic: Powered by SQLGlot, it transpiles the semantic queries natively to Postgres, Snowflake, Databricks, BigQuery, etc.

• The Built-in Explorer UI: I didn't want it to just be a headless CLI. If you run pysemantic studio, it spins up a local Streamlit dashboard where you can visually debug your entity graph and test query generation in real-time.

Check it out here:

• Docs: https://pysemantic.com/

• GitHub: https://github.com/iamniteshjangir/semantic-py (Drop a star if you find it useful!)

• Quickstart: pip install pysemantic-layer

I’m currently exploring ideas like "Semantic FinOps" and adding a semantic layer for AI agents, but right now I am focused on making the core engine bulletproof.

I’d love for you to try it out, poke holes in my architecture, or tell me where the query planner breaks. Brutal feedbacks are welcome!