Open-source tool to run LLM models on DWH data
Hi folks, this is Burak, co-founder at Bruin. We built an open-source framework to run inference workloads on your DWH data, e.g. Snowflake, Databricks or BigQuery, and enrich the data using an LLM.
There are a lot of cases where you have some data that lives inside Snowflake, and you want to enrich it using language models. This might be a call to a Haiku model to check if a ticket needs attention, or Typesafe's Jev to categorize the issue.
Good luck if you want to do this inside your data pipeline. You will either have to run your dbt pipelines first and then do the enhancement, essentially splitting your pipelines and dependencies, or have to build a brand new Airflow pipeline.
I think that should have been a better experience.
We are making inference a first-class primitive inside Bruin pipelines today. They allow bringing language models into your data pipelines. You can run models on every row in any table and produce additional tables with new columns. You can use Claude, GPT, Jev, or any other model on the planet.
Point your inference asset to a table, give the prompts for each column to be generated, and watch Bruin fill out all the columns you have as efficiently as possible. It takes care of compatibility across providers, preserving caches, and batching requests where possible.
This is all open-source and available today. You can point your agent to our docs and tell it to build a quick inference pipeline over your Snowflake/Databricks/BigQuery tables.
bruin-data/bruin
getbruin.com
Super excited to launch this, and I'd love to hear what you all think!
1 point · 5 comments
Mentions
Open-source tool to run LLM models on DWH data Hi folks, this is Burak, co-founder at Bruin. We built an open-source framework to run inference workloads on your DWH data, e.g. Snowflake, Databricks or BigQuery, and enrich the data using an LLM. There are a lot of cases where you have some data that lives inside Snowflake, and you want to enrich it using language models. This might be a call to a Haiku model to check if a ticket needs attention, or Typesafe's Jev to categorize the issue. Good luck if you want to do this inside your data pipeline. You will either have to run your dbt pipelines first and then do the enhancement, essentially splitting your pipelines and dependencies, or have to build a brand new Airflow pipeline. I think that should have been a better experience. We are making inference a first-class primitive inside Bruin pipelines today. They allow bringing language models into your data pipelines. You can run models on every row in any table and produce additional tables with new columns. You can use Claude, GPT, Jev, or any other model on the planet. Point your inference asset to a table, give the prompts for each column to be generated, and watch Bruin fill out all the columns you have as efficiently as possible. It takes care of compatibility across providers, preserving caches, and batching requests where possible. This is all open-source and available today. You can point your agent to our docs and tell it to build a quick inference pipeline over your Snowflake/Databricks/BigQuery tables. bruin-data/bruin getbruin.com Super excited to launch this, and I'd love to hear what you all think!