Dune dashboards are a potent C2C tool for getting complex, updated crypto analytics. It uses SQL language to query the data for different visualizations. However, many community members had never heard of the SQL language. Furthermore, the more complex information you want to visualize, the trickier it gets to describe it via SQL. It is easy to get confused after a couple of joins on top of other joins. We utilized the latest achievements in the NLP field to lower this entry threshold and make this tool available and accessible to anyone who wants to get insights and share them with the community. All you need to do is describe what kind of data you want to visualize and what transformations you want to perform on it. AI will then translate this to SQL query, and you will be able to create and share your analytics ideas with the community! The power of AI will enrich what the platform could suggest to the community.
⁃ We had to use OpenAI's API to access the GPT-3 network. There's not enough time and resources to train such a Large network during the weekend
⁃ Data collection. No Dune's API endpoints allow us to download public queries, so we had to scrape Dune's web to collect the dataset.
⁃ Data labeling. Every scraped query must be labeled with natural language to fine-tune the GPT-3 model for this task. We have decided to auto-label the data since there was no time to do it manually.
⁃ Fine-tuning the model. We had a lack of time and resources to get accurate results.