I have frequently encountered challenges with the non-standardized responses from ChatGPT when seeking assistance with writing custom functions, actions, or widget code for FlutterFlow. Despite eventually generating the required code through multiple prompt corrections, the lack of direct integration between FlutterFlow documentation and ChatGPT's vector database often resulted in accuracy gaps.
To address this issue, I devised a solution involving the following steps:
scrape the https://docs.flutterflow.io/
Install AnythingLLM on Windows: I downloaded and installed AnythingLLM, which serves as a versatile language model manager.
Download and install https://lmstudio.ai/ (you can use ollama and others, but i wanted better control and was planning to use lmstudio embedding model)
Integrate Documentation into AnythingLLM: Instead of using a Google Custom Search Engine, I fed the scraped FlutterFlow documentation into AnythingLLM in PDF format. This direct integration ensures more accurate and relevant responses.
Configure a System Prompt: I created a well-defined system prompt to guide the model in providing precise and relevant assistance.
Note: Speed of text generation, code generation and overall response depends upon your system specifications, choosing correct model for lmstudio.
Depending upon your need switch between chat & query in workspace settings in AnythingLLM
