Back office ops · Production

Circuitry.ai achieves 60–70% reduction in information search time with Databricks RAG chatbots

The problem

Circuitry.ai's small technical team faced delays building RAG chatbots due to challenges applying metadata filters on retrievers, maintaining knowledge base updates without disrupting RAG chains, ensuring proper data segregation of proprietary customer data, and integrating multiple data sources with differing structures and formats.

Workflow diagram · grounded in source
1
Customer document upload
trigger
“The workflow began when customers uploaded documents via the user interface, triggering three key tasks”
2
Data ingestion and embeddings
ai_action
“triggering three key tasks: ingesting raw data, processing and generating embeddings and deploying the serving endpoint. Each task was supported by associated notebooks handling required inputs and dependencies from upstream tasks”
3
Endpoint published to app stack
integration
“Once the endpoint was ready, it was published to the application stack for customer use”
4
RAG pipeline query answering
ai_action
“Circuitry.ai implemented a RAG pipeline to enhance the accuracy and relevance of the answers provided by their AI chatbot tools. This pipeline utilized generative AI models, specifically Llama and DBRX, which are known for their precisio…”
5
User feedback for improvement
feedback_loop
“incorporated a feedback mechanism to ensure continuous improvement, where users could rate the GenAI-generated responses. For example, this feedback loop helped Circuitry.ai improve their Decision AIdvisor tool by simplifying the queryin…”
Reported outcome

Customers of Circuitry.ai's decision intelligence software experienced a 60–70% reduction in time spent searching for information, with AI responses delivered in seconds rather than minutes and overwhelmingly positive feedback from proof-of-concept trials.

Reported metrics
Time spent searching for information60–70%
Response delivery timeanswers in seconds rather than minutes
customer feedback from PoC trialsoverwhelmingly positive
Reported stack
Databricks Data Intelligence PlatformDelta LakeUnity CatalogMLflowModel ServingLlamaDBRX
Source
https://www.databricks.com/customers/circuitry-ai
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Customers of Circuitry.ai's decision intelligence software experienced a 60–70% reduction in time spent searching for information, with AI responses delivered in seconds rather than minutes and overwhelmingly positive…

What tools did this team use?

Databricks Data Intelligence Platform, Delta Lake, Unity Catalog, MLflow, Model Serving, Llama, DBRX.

What results were reported?

Time spent searching for information: 60–70%; Response delivery time: answers in seconds rather than minutes; customer feedback from PoC trials: overwhelmingly positive (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Customer document upload → Data ingestion and embeddings → Endpoint published to app stack → RAG pipeline query answering → User feedback for improvement.

Related back office ops cases