Today I won't be posting the daily market overview... Over the weekend, I will share this week's on-chain analysis..

Recently, I've hit some snags in handling these tens of thousands of research reports... It all started with a fairly grand idea, aiming for a structure like this: Natural language demand -> LLM breaks down the demand -> LLM fetches data -> validation -> fetch more if needed -> start analysis after fetching -> finish analysis and begin writing, and so on..
The initial version was a manually crafted plan → Act → Observe → Reflect state machine.. I also tried LangGraph, and found that while it can run, the results are mediocre..
It's just a simple demo.
So from the start, I felt that trying to go big was hitting the boundaries of vibe coding that I can't get past for now..
A framework that's too large makes comprehensive optimization difficult...
Now I am returning to small tasks, going back to the previous approach... First, I’ll build a bunch of small workflows... For example, summarizing over 300 research reports updated today, identifying the most worth reading... The overall idea, market sentiment, where are today's research report hotspots, are there any directions that need further research today, are there any stocks worth pitching, etc...
Gradually expanding to multi-day deep dives into individual stocks/industries...
Starting from hundreds of small workflows daily, slowly expanding to thousands or tens of thousands of workflows...
And then gradually integrating these small workflows into a larger framework...
That should be a better approach..