The CLUE Technique: Grounding Generative AI

 

The CLUE technique offers a systematic approach to preserving and reusing database context, enabling more efficient and accurate interactions with AI. With the CLUE technique, you can build a reusable starting point for your database conversations.

C: Continuing Established Conversations

The first element of the CLUE technique leverages the AI's ability to maintain context throughout a conversation. Instead of starting fresh each time you need to work with your database, create dedicated conversation threads organized by analytical purpose.

Imagine you've created separate threads for inventory analysis, customer behavior reporting, and sales performance tracking. Within each thread, you begin by thoroughly establishing your database context—explaining your database structure, relationships, and providing sample data. After asking your initial questions, you can return to this same conversation days or weeks later for related queries.

To keep these ongoing conversations manageable, periodically summarize what you've established or add clear markers like "DATABASE CONTEXT ENDS HERE" to help you quickly locate important reference points when returning after a break. If a conversation becomes unwieldy or is using up your quota for your LLM provider due to length, create a new one but copy the essential context-setting messages from the previous thread.

L: Leveraging Existing Documentation

Most organizations already have documentation about their databases that can be repurposed for AI context, forming the second component of the CLUE approach. You can easily attach these documents to prompts in most LLMs. Look for a "+" or paper clip icon next to where you type in prompts.

Look around for resources like data dictionaries, entity-relationship diagrams, database design documentation, and data governance materials. These often contain precisely the information needed to establish proper context for AI interactions, saving you from creating new materials from scratch.

When uploading such documentation, if you don't need it all, provide guidance about where to find key information: "I'm uploading our official data dictionary. Pages 3-5 contain table definitions, pages 6-8 detail the relationships between tables, and page 12 has our data formatting standards." Adding this navigational help ensures the AI focuses on the most relevant sections of sometimes-lengthy documentation.

U: Uploading Context Documents

The third element of CLUE involves creating and uploading comprehensive context documents specifically designed for AI interactions. Rather than retyping everything from scratch, just create a shared document that has the essential contextual information and reuse it when you start new conversations.

Your context document should include detailed database environment information such as the database type and version you're using, your connection client, and connection method. Include schema information with your database name and a complete listing of tables with their columns and data types. Map out table relationships to help the AI understand how to properly join tables in queries. Perhaps most importantly, include representative data samples that illustrate typical patterns and formatting conventions.

When starting a new conversation, upload your context document with a prompt that directs attention to it: "I'm attaching information about my database structure. Please reference this throughout our conversation." Before proceeding with specific questions, confirm that the AI has understood the context by asking it to summarize what it knows about your database environment.

E: Exemplifying with Sample Queries

The final element of the CLUE technique leverages your existing queries as concrete examples of how your organization interacts with the data. You probably already have lots of queries you are running. If so, give a bunch of them to the AI.

Sample queries provide powerful context about your database structure, naming conventions, and query patterns. When selecting representative examples to share, choose queries that cover different aspects of your database, demonstrate typical joining patterns, show how you handle common data transformations, and illustrate your SQL style preferences.

For each query you share, include the actual SQL code along with a brief explanation of what it accomplishes, any notable techniques used, and common modifications you typically make. By seeing how you join tables, which fields you commonly filter on, and how you structure complex calculations, the AI gains deeper insight into both your database structure and your analytical approach.

Putting CLUE into Practice

The power of the CLUE technique comes from combining these approaches into a seamless workflow. Here's how that might look in practice:

1. Create a comprehensive context document that includes database environment details, schema information, sample data, and common sample queries with explanations.
2. Start a new conversation by uploading this document and confirming understanding.
3. Continue this conversation for related queries over time, leveraging the established context.
4. When exploring a significantly different area of your database, incorporate existing documentation relevant to that domain.

By implementing the CLUE technique, you can dramatically reduce the time spent establishing context, minimize the risk of hallucinations in AI-generated queries, and build an increasingly accurate understanding of your specific database environment. This transforms generative AI from a simple query suggestion tool into a collaborative data exploration partner that understands the nuances of your database and analytical needs.