Need for an Integrated Solution in Vanna Without Relying on VannaFlaskApp

[ruben-quilez]

Hello everyone,

I'm working with Vanna and have encountered a difficulty that I'd like to share and discuss with the community.

While it is technically possible to use Vanna's base functions to implement a complete workflow that includes:

• Training with table schemas and documentation.
• Providing feedback and maintaining memory.
• Query rewriting.
• Interacting with responses and generating summaries.
• Debugging and understanding the reasoning behind queries.

...without relying on VannaFlaskApp, I've realized that doing so involves a significant amount of work to recreate all the orchestration logic and management already present in VannaFlaskApp.

My concern is more a matter of software architecture and design. The question is:

• Why do we have to reinvent all this infrastructure when it could come integrated, as it does with other tools like Qdrant?

By having to implement all the orchestration logic ourselves, we face several challenges:

1. Maintenance and Updates: Every time Vanna updates, we'll have to review and possibly update our entire implementation to maintain compatibility and take advantage of new features.

2. Focus on Business Logic: Instead of focusing on developing our business logic, we spend time and resources rebuilding infrastructure that already exists in VannaFlaskApp.

3. Complexity of Implementation:

   • Complex Interactions: The base functions are highly interrelated. For example, an action like ask() may involve calls to generate_sql(), get_similar_question_sql(), get_related_ddl(), among others.
   • State Management: It's necessary to handle the state between multiple calls, coordinate the feedback loop, manage auto-training, and maintain a coherent history.
   • Handling Edge Cases and Errors: Implementing all the checks and error handling already covered in VannaFlaskApp is a considerable task.

To illustrate with an analogy:

• It's as if Qdrant only provided us with basic functions and told us 'now build your own vector management system'. Could it be done? Yes. Does it make sense? Not really, because it implies duplicating efforts and distracting us from our main goal.

My question is:

• Is there any way or plan to use Vanna in a more integrated manner without relying on VannaFlaskApp, but without having to reinvent all the infrastructure necessary for a complete workflow?

• Are there plans to provide a more integrated solution that facilitates implementation and maintenance, similar to how other tools like Qdrant offer more complete infrastructure from the start?

I understand that VannaFlaskApp provides more than just an interface; it handles much of the logic and state management. However, my goal is to achieve the flexibility to customize the workflow without losing efficiency or having to rebuild all the underlying infrastructure.

I would greatly appreciate any guidance, information on future plans, or advice on how to approach this challenge. My intention is to make the most of Vanna, focusing on business logic and avoiding unnecessary efforts on the underlying infrastructure.

Thank you very much for your time and help!

---

Note: I'm not simply looking to know if it's possible to implement the workflow without VannaFlaskApp (I know it is), but to discuss whether there are more efficient ways to do it or if there are plans to offer a more integrated solution that avoids having to reinvent the wheel.`