Hi,
I'm wondering if the team could share their AI context rules so I could apply Wappler knowledge to AI tools outside the Wappler editor?
I'm poking around with Claude Code after seeing a fellow Wappler used it to create a tool for Wappler...
Best thing to do is provide examples of what you want it to do. Train the LLM you are using on these examples and tweak and provide further examples advancing the complexity as you progress. Have the LLM store its knowledge to a text file every time it does something right and also when it does something wrong. Create an example Project and then open your prefered IDE and direct it to the example Project. Ask it to describe each aspect of the Project within a text file. Correct it along the way step by step. Rinse and repeat dozens of times. Be prepared for the worst before you see the best results. Have one Modal interpret the results of another and each review the others understanding to grow the data set used for training. Then do it all again. Explain what it is you have done to the LLM. Provide very concise rules in a rules.txt file. Ask the LLM to interpret these rules and improve upon them then have another LLM do the same, and even a third, fourth and fifth. Then go back and ask the first one to review them again. Instruct any LLM to always follow the rules. Document success and mistakes. You really have to drill it in to these LLMs to follow rules otherwise they'll spin off out of control and repeat the same mistakes. Always keep a chat log file that documents your conversations and have the LLM refer to this prior to picking up where you left off. One of the first rules should be to always refer to the chat log and always update it periodically with a concise synopsis of what you have discussed within the session. All LLM's forget what they have done some in more spectacular ways than others. I've had Claude forget what it did minutes before so be vigilant and watch their output and stop them when you see them making a mistake and tell them. During today's Meet Up we discussed this for a while. Good input good output or as it was phrased another way garbage in garbage out. It really is a lot of repeat and improve and the more times you do that the better things become. AND always backup everything so have the LLM create an automated backup script that runs every time a file is changed and have it store these backups in to a time stamped directory. Then once again outline this in your rules file as a clear instruction to the LLM. Not exactly what you are asking for but hopefully helpful none the less...
1 Like