Interesting, I'll give it a try. I've just finished a project that might benefit from this framework. Thanks for sharing.

Hey @Angel Castro , If I were to build this from scratch I would focus on learning every part of the solution. You could take the lovable path but that takes out the learning experience and when problems arise only the people that knows how each part works and connect will be able to fix it or enhance it. Topics to think about: - Core architecture and tech stack: I would use python, an LLM service for text generation, an embedding model for semantic search and retrieval, postgreSQL for storing data, and pgVector extension for vectorisation. - Brand voice: for this you'll need data in the form of blogs, articles, reports, etc to capture your style. They will be stored in a vector database and retrieved. - After setting up a knowledge database and testing the content generation workflow, include SEO & GEO capabilities and image generation into the generation workflow. - Develop or use a REST API integration software to publish content into workpress. - Deploy and scale. This is just a high level dump of things I would plan initially before starting to work on this.

@Enzo Piñeyro. You could start with python frameworks for web applications. They are normally wrappers and encapsulate most of the html/css/javascript learning curve challenges into functions. If you are starting out and want to get some UI developed, have a look at: Dash, Streamlit or Shiny for Python. I've tried Dash and Shiny for python and they are very similar. Haven't tried Streamlit yet. They all started as dashboarding tools but the features they available today can give you enough for a web application. I myself have used them in several web apps. Good luck.

If you depend on LLMs applications to do the coding for you then your level of expertise is determined by how good LLMs are in coding, otherwise, if you know how to code then your expertise is exponential as you can use LLMs to do bits of code but you put them all together into solving a big problem. There is nothing wrong with just using prompts to solve problems, but eventually you will have to have someone to maintain that code when things break or want it expanded to another feature. If you want to stay a generalist then learn to identify when AI is a good tool for a problem, no code needed at that level, but good decision making.

hey @Kien Hua. Been developing analytics systems for sometime so I'm going to give you my two cents, not the truth but my learnings from experience: - Treat LLMs as a tool to an end, not the end itself. Same with AI, machine learning, etc. - Stakeholders get attracted by the shiny ball very easily so their requests normally come in the form of "use AI...", "can we use LLM to solve X...", etc. - Requirements for any software project change over time, because people don't know what they want until they see something, so ellicit their first idea of what they need to solve and give them a prototype to adjust their thinking/assumptions. - the best to ellicit good requirements is to ask better questions. Good questions make the mind focus. So, instead of "what do you need?" say "what are you trying to accomplish?" or "once you get this data/info, what decision(s) are you making?". Instead of "what chart do you want to have here?" say "why is this data important to you?". - From the first conversation, break down the project into features and then analyse which features can be developed with LLMs, AI, machine learning or statistics only. - Develop a few features that tackle the core problem and show it to them. If that's in the right direction continue, otherwise, pivot, build, show and continue. That's in a nutshell, there are a lot little details that I ommitted here due to space and time, but I'm happy to continue the conversation in the thread :)