"The test results are out, there is no trace of artificial addition, it is all AI" Previously, about 95% of the Dune Query was done by AI. This article focuses on DuneAI and how to use AI to efficiently engineer prompt words (AI intermodulation?). ） Part 1: A brief introduction to using Dune AI @Dune built-in AI helps users quickly generate SQL queries using natural language. After creating a new query, you can see the prompt input slot at the bottom of the code block Points of use: 1. Clearly express what you want to query in English, and provide context, including key information such as database, time range, specific contract address, token symbol, etc - Example: "List all trades involving WETH (0xC02aaA39b223FE8D0A0e5C4F27eAD9083C756Cc2) on Uniswap v3 in the past 24 hours" 2. Adjust your query requirements based on the initial results - Dune AI has built-in test cases to ensure that the generated queries work correctly - You can use the "fix" function at the top of the query code to modify and fine-tune based on existing queries For more information, please refer to the official documentation: Part II: Multi-Agent Prompt Word Project In practice, relying on one AI tool alone often struggles to get the best results. The set of multi-agent collaborative Prompt Engineerimg processes introduced here can significantly improve query quality and efficiency, and can be extended to other business scenarios Basic workflow 1. Mission Description: Start by articulating the business requirements 2. AI1 processing: Use general-purpose AI, such as ChatGPT or DeepSeek, to translate requirements into professional prompt words 3. AI2 execution: Enter the optimized prompt words into a professional AI (such as Dune AI) for execution 4. Outcome Evaluation: Check whether the preliminary results meet the needs 5. Iterative optimization: Adjust the prompt words as needed and repeat the above process You can flexibly choose AI2 that is close to your business scenarios based on your needs, such as using Dune AI to generate queries, Copilot Claude 3 to complete programming tasks, and Grok 3 to retrieve and organize Twitter information For example, my first prompt is: “ I need to analyze the trading behavior of a specific token on the Solana chain to find out what is going on within the specified time frame: 1. Addresses that have bought more than N times 2. Addresses with a buy-in amount greater than X USD 3. Addresses with net inflows (total buys - total sells) greater than USD Y 4. Addresses that have not been sold more than Z times Eventually, the details of these addresses will need to be returned, including: - Wallet address - Net inflow amount - Number of buy-ins - Number of sells - Time of first trade - Last trading time - Token address Please help me translate this requirement into a professional prompt word format that Dune AI can understand, including all necessary schema and parameter definitions; Use English; Related database documentation: (I have provided a description in here) ” After that, I fine-tuned the second stage prompt and let Dune AI generate the query; I put the second stage prompt I used in the comment section Hope this article was helpful to you (Yes, this trick is also used in this article, and the images are all drawn with AI after generating the mermaid syntax)