How AI Powers an Immersive Vibe Analyzing Experience for Data Exploration

1. What Is AskTable

AskTable was created to close the gap between data and insight by moving analysts from a consensus‑efficiency stage to a cognitive‑increment stage, where they can proactively detect anomalies, reveal trends, and support strategic decisions. The platform evolved from a 2023 "ChatBot for Database" prototype that proved natural‑language querying feasible, to a 2024 configurable Agent platform, and in 2025 split into two core products: the Question Engine for instant, accurate self‑service queries and the Analysis Canvas for an immersive, collaborative analysis environment. Together they form a new AI‑driven data‑analysis paradigm described as "infrastructure + application".

2. Question Engine

The Question Engine is the foundational infrastructure that supports intelligent, high‑performance queries across heterogeneous sources such as MySQL, Doris, Excel files, and API endpoints. Users write natural‑language questions, which the Table Agent Builder translates into precise SQL statements that the system executes. Results are returned to business users via web pages, apps, Enterprise WeChat, Feishu, DingTalk, or custom OA/CRM systems, delivering a seamless "query directly in the business scenario" experience. By eliminating pre‑built models and fixed dashboards, the engine reduces latency and removes the need for IT scheduling. However, because results are generated dynamically, traditional page‑level permission controls are insufficient; AskTable moves permission enforcement to the data‑source layer, applying row‑ and column‑level policies based on the original tables.

3. Analysis Canvas

While the Question Engine handles simple look‑ups, deep analysis requires divergent exploration, which pure chat‑based BI (ChatBI) cannot support. AskTable therefore proposes a "NoChatBI" approach that productizes, visualizes, and immerses the analysis process. The Canvas splits the workflow into two roles:

Builder : Guided by natural‑language commands, AI automatically generates SQL, Python code for cleaning, aggregation, and machine‑learning, and then uses JavaScript to render visualizations, enabling low‑barrier multi‑source data fusion and complex computation.

Viewer : Observes the AI‑generated reasoning chain, summaries, and insights, helping users understand the analysis path and accelerate report interpretation.

On the two‑dimensional canvas, users can drag data, define relationships, iterate analyses, and finally export reports or dashboards. This design replaces traditional ETL and coding pipelines with a seamless transition from thought to expression, allowing analysts to co‑create with AI efficiently.

The Canvas offers three core functions:

Natural‑language query → SQL nodes that are traceable and editable.

AI‑driven chart recommendation with interactive adjustments.

Cross‑source data processing via generated Python scripts for field calculations, cleaning, and multi‑source merging.

4. AI Agent Technical Concept

Data alone cannot deliver complete insight; business logic embedded in code or APIs must be combined with structured facts. AskTable’s "Any to Table" pipeline converts unstructured sources (web pages, CRM records, chat logs) into analyzable tables, then merges them with external APIs. The Vibe Analyzing technology fuses data and logic to generate actionable insights, turning "raw data" into "decision‑support" outcomes.

5. Accuracy Challenges and Mitigations

Accurate AI‑driven analysis requires joint business‑technical effort. AskTable adopts a GenSQL strategy that explicitly distinguishes known from unknown to avoid hallucinations, couples high‑quality large models with schema linking for better table‑relationship understanding, and employs test‑driven model evaluation. Best practices include limiting the query scope (e.g., to 50 tables), providing complete context, avoiding overly complex tasks, and validating models before deployment.

6. Decomposing Tasks with AI Canvas

The Canvas reuses the engine’s capabilities, breaking complex tasks into atomic agents via a Workflow+ approach. This yields better controllability than fully autonomous Agentic AI, supports transparent debugging (viewing generated SQL/Python), and enables a human‑in‑the‑loop process that improves reliability.

7. Table‑Finding Problem

Locating the correct table among many is a key challenge. AskTable adopts an adaptive strategy based on table count:

Natural‑language query for very small tables (high efficiency).

Direct large‑model inference for few tables.

RAG (retrieval‑augmented generation) with vector‑based recall of table schemas for medium‑scale scenarios.

Agentic AI with multi‑round trial‑and‑error for large‑scale environments.

This multi‑mode approach balances inference, retrieval, and autonomous exploration to achieve precise and efficient table selection.

8. Data Access Permission Control

Two mechanisms ensure fine‑grained security:

RolePlay : AI identifies the user’s role and applies dynamic, granular policies (e.g., store‑level managers can only view their own store data).

Row/Column Control : After AI generates SQL, a classic database permission engine parses the statement and enforces row‑ and column‑level filters to guarantee that output complies with security requirements.

9. Best Practices and Case Studies

AskTable has been deployed in state‑owned enterprises, financial institutions, and cloud‑service providers. Notable use cases include:

China Construction: Integrated into an internal AI platform to enable natural‑language queries for project‑management systems.

Guoyuan Securities: Embedded in the R&D workflow to auto‑generate SQL and explain table structures, reducing developers’ data‑understanding cost.

Kingsoft Cloud: Replaced cumbersome report look‑ups with an AI assistant that quickly retrieves reimbursement, customer‑consumption, and performance metrics.

Medical, retail, and education sectors: Integrated AskTable into enterprise WeChat, API‑driven mini‑programs, and knowledge‑base platforms to provide instant, natural‑language data access across diverse domains.

10. Q&A Highlights

In the Q&A, the team emphasized that the core of AskTable is a large‑model‑driven SQL generator, with schema linking used only as contextual aid. The product remains lightweight, generic, and standardized, allowing the same architecture to serve multiple industries without deep customization.