Supply chain · Production

Altana deploys GenAI models 20x faster with up to 50% improved performance on Databricks Mosaic AI

The problem

Before Databricks, Altana's team was forced to build and maintain boilerplate GenAI launch infrastructure and evaluation tooling instead of focusing on product functionality, while managing competing tensions between vendor lock-in, total cost of ownership, and information security across multiple cloud deployments.

First attempt

Altana's prior infrastructure offered no unified ML lifecycle platform, requiring engineers to create their own boilerplate tooling for every GenAI launch and leaving them unable to resolve competing demands around vendor lock-in, cost, and security without sacrificing product velocity.

Workflow diagram · grounded in source

Delta Lake data ingestion

integration

“Delta Lake, which offered real-time data ingestion and robust data management capabilities”

Custom deep learning (early phases)

ai_action

“custom deep learning models for the early phases”

RAG-based agent workflow refinement

ai_action

“fine-tuned agent workflows with retrieval augmented generation (RAG) for later refinements”

Customer RLHF feedback loop

feedback_loop

“reinforcement learning from human feedback (RLHF) from customers is used for further model refinement”

MLflow model evaluation

validation

“model and LLM evaluation tools in MLflow, combined with Model Serving. These have allowed our ML teams to rapidly train, test and deploy their own models for customers”

Model serving and deployment

output

“We're training and deploying models more than 20 times faster to production”

Reported outcome

After migrating GenAI development workflows to Databricks, Altana trains and deploys models more than 20 times faster to production and has seen model performance improve by 20–50%, freeing the team to focus on building AI-driven products rather than maintaining infrastructure.

Reported metrics

Model deployment speed20 times faster

Model performance improvementup to 50%

Model performance improvement range20–50%

Reported stack

DatabricksDatabricks Mosaic AIManaged MLflowModel ServingDelta LakeUnity CatalogDBRXLlama 3Phi-3RAGRLHF

Source

https://www.databricks.com/customers/altana-ai

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After migrating GenAI development workflows to Databricks, Altana trains and deploys models more than 20 times faster to production and has seen model performance improve by 20–50%, freeing the team to focus on buildi…

What tools did this team use?

Databricks, Databricks Mosaic AI, Managed MLflow, Model Serving, Delta Lake, Unity Catalog, DBRX, Llama 3, Phi-3, RAG.

What results were reported?

Model deployment speed: 20 times faster; Model performance improvement: up to 50%; Model performance improvement range: 20–50% (source-reported, not independently verified).

What failed first in this deployment?

How is this supply chain AI workflow structured?

Delta Lake data ingestion → Custom deep learning (early phases) → RAG-based agent workflow refinement → Customer RLHF feedback loop → MLflow model evaluation → Model serving and deployment.