Struggling to help junior analysts get data from your SQL platform without relying on experts? Looking to accelerate adoption of your SQL environment for less-technical users?
Large Language Models (LLMs) can now generate SQL for analytical databases, but they need context, control, and precision to be useful in real-world environments.
SQL experts can pull data in seconds. But for non-SQL-experts, like junior analysts or technical leaders, the same task may take 30 minutes or more, or they escalate to someone else, increasing workload and adding delays.
The answer: Let users ask questions in plain English, and get the data they need, fast without being SQL experts.
Yellowbrick AI Capabilities for Enterprise Analytics
Yellowbrick can turn natural language into reliable, secure, and accurate SQL queries, through a series of Artificial Intelligence capabilities to bring speed, security, and intelligence to enterprise analytics.
- Text-to-SQL: Ask questions of their data in plain English and generate the equivalent SQL instantly from an LLM.
- Bring Your Own LLM: Connect to private LLMs (on-premises or private cloud), or cloud services from OpenAI and others, while keeping full control of your data.
- Vector Store for RAG applications: High-performance SQL vector search, enabling Retrieval-Augmented Generation (RAG) to pull data from company-specific context for better answers, while the team keeps using familiar SQL tools.
How do these AI capabilities work together?
1. Translating Natural Language to SQL with Context
Ask a question like: “What was the total rainfall in inches in CARY, NY, in January 2025?”
The LLM generates a complete SQL query that: joins the relevant tables, filters by period and region, and applies the appropriate logic and formatting for your schema.
2. How the LLM Builds the Query
Behind the scenes, the prompt sent to the LLM includes:
• The schema for the relevant tables (e.g., observations, tables by state, year, etc).
• Sample values from each column to guide formatting and filters.
• An instruction to behave as an SQL expert.
This structure allows the LLM to generate precise, executable SQL aligned with your actual schema and data patterns.
3. Vector Search for Schema Relevance
When schemas scale into hundreds or thousands of tables, manually picking the right tables becomes unmanageable. Yellowbrick uses vector similarity search to match the user’s question to the most relevant tables, optimizing the context sent to the LLM.
4. Enforce Access and Governance
Admins lock down outbound API calls and whitelist approved LLM endpoints. This gives enterprises the ability to enforce strict data egress policies without compromising usability.
How Configurable Is Yellowbrick’s Text-to-SQL?
Yellowbrick’s copilot provides granular control over how prompts are built, and which models are used:
• Model selection: Works with GPT-4o, DeepSeek, or any OpenAI-compatible model, including local deployments.
• Prompt value limits: Configure the number of sample values per column to manage token count and prevent data leakage.
• Vector search: Enable automatic table selection using semantic similarity across thousands of tables, no manual curation required.
• Embedding model choice: Set which embedding model powers table relevance scoring.
• Prompt profiles: Create reusable templates tailored to specific roles or query types.
• Security controls: Admins can restrict outbound LLM traffic to approved domains, ensuring compliance with data governance policies.
What are the business benefits?
• Reduce skill barriers: New or junior data engineers can get answers without deep SQL knowledge.
• Simplified architecture: No separate vector database needed.
• Make faster decisions: AI answers are grounded in your proprietary, governed data.
• Stay secure: Sensitive data stays inside your organization.
• Support compliance: Aligns with industry and legal requirements.
With Yellowbrick AI, your teams move faster, your data stays secure, and your business makes decisions at the speed it needs.
Try It Yourself
Want to test this in your own environment?
Try Yellowbrick for free and let your data team ask questions the way they think, and get back answers in seconds, using SQL.
