Vanna's Vision & Roadmap

Vision Statement

Vanna aims to redefine the landscape of data analysis by becoming the default tool for creating an AI data analyst. We envision an AI that approaches the capabilities of a human data analyst in terms of accuracy, interactions, and autonomy.

Our journey begins with mastering SQL generation through the use of Large Language Models (LLMs) because:

• SQL has a high token density, making it an ideal target for context-limited LLMs.

• SQL is the most common ad-hoc interface into data.

• A lot of value is locked inside SQL databases that only a subset of users at organizations can access.

Roadmap to Realizing the Vision

Our roadmap is structured around three pivotal goals: Accuracy, Interactions, and Autonomy. These goals are designed to ensure that Vanna's AI data analyst not only meets but exceeds the expectations of our users, paving the way for a new era of data analysis.

1. Accuracy

Objective: Achieve unparalleled accuracy in SQL generation and data analysis, mirroring the expertise of a seasoned data analyst.

Notes:

• Instead of benchmarks, we should focus on "experienced accuracy."

• This is already largerly achieved by storing historical correct question-SQL pairs and using them to inform the LLM's responses.

• Databases evolve over time, so we need to be able to "manage" the AI's knowledge base just as one would manage a human's knowledge

2. Interactions

Objective: The AI should interact with users in a way that is familiar and collaborative, similar to how a human data analyst would interact with stakeholders.

Notes:

• If there's information that the AI doesn't have, it should be able to ask for it

• If the question is ambiguous, the AI should be able to ask for clarification

• The AI should be able to provide explanations for its answers

• The AI should be able to provide a confidence score for its answers

• The AI should suggest follow-up questions

3. Autonomy

Objective: Empower the AI to function with a high degree of autonomy, capable of accessing common systems and performing downstream actions based on its analyses.

Notes:

• Streamline the onboarding process to give the AI immediate access to the necessary systems and data sources.

• Enable the AI to autonomously execute follow-up actions based on its data analysis, such as generating reports, alerting stakeholders, or triggering workflows.

Open Source

We believe that the best way to achieve our vision is for Vanna to be open source. This will allow the community to contribute to the project and help us achieve our goals faster.

By having a modular design, we can meet any organization's security requirements since you can plug it into components that are already approved by your security team.

Development

All of our development efforts should be aligned with the vision and roadmap.

(Living Document, Last Updated 2024-02-27)