Automatic Chart Selection & Multiple Visualizations
DatAsk automatically selects the most suitable chart and supports multiple chart types.
AI Insights for the Chart
Flexible Dashboard Customization
Custom Dashboard Example (DatAsk)
One-Click Dashboard Publishing
Shareable Public Dashboard
Transparent SQL & Data Access

DatAsk — AI for Internal Reporting & BI

Inspiration

In every enterprise, managers are making decisions without data. Not because the data doesn't exist — it's sitting in the warehouse — but because nobody has time to pull it out.

A branch manager needs yesterday's transaction numbers for the 9am meeting. They email the ICT reporting team. By 2pm the report arrives. But the meeting ended at 10am, and the decision was already made on a gut estimate. Multiply this by hundreds of ad-hoc requests every week and you get two painful realities: an ICT team drowning in MIS tickets, and business leaders running the bank on stale data.

Meanwhile, Power BI and Tableau exist — but they require IT to model the data, business users to learn DAX, and a drag-and-drop ritual just to answer a simple question. For the 99% of employees who aren't analysts, self-service BI has been a promise, never a reality.

We set out to build something different: a BI layer that speaks natural language, in Vietnamese, with the auditability and governance a bank actually needs. No dashboards to configure. No DAX. Just ask.

What We Built

DatAsk is a fully AI-powered self-service BI platform featuring:

Conversational Query — Ask any question in natural Vietnamese or English; Qwen3-Max generates schema-aware SQL and returns the answer in seconds
Auto-rendered charts — The right visualization (bar, line, pie, number) is picked automatically based on the shape of the data
Artifact Dashboard — Every answer is saved as a reusable artifact; drag and drop them onto a canvas to compose a dashboard in minutes
One-click Publish & Share — Generate a time-bound, read-only link so colleagues can view dashboards without logging in
Human-in-the-loop governance — Generated SQL is always visible, raw data is always shown, AI narrative insights are explicitly labeled and manually triggered

Prompt-to-Dashboard: Empowering Non-Technical Users

The game-changer of our solution: anyone can build a dashboard without touching a single field mapping, DAX formula, or query builder.

A branch manager simply types:

"Show me total transactions by day for March 2026, and highlight the branches with the biggest month-over-month growth."

Qwen3-Max instantly:

Inspects the connected schema (customers, transactions, branches, products)
Generates a complete SQL query — joins, aggregations, window functions, all correct
Executes the query against the read-only data warehouse
Picks the optimal chart type (bar chart for time series, horizontal bar for ranking)
Writes a plain-language AI Insight summarizing what the chart shows
Saves everything as an Artifact — re-runnable, auditable, shareable

The user can then drag the artifact onto the Dashboard Canvas next to other saved answers, resize widgets, hit Publish, and send a link. From "I have a question" to "my team has a live dashboard" — under 5 minutes.

Why this matters for banks and enterprises

Zero engineering cost per report — Every department gets self-service analytics without adding BI headcount
Business teams own their dashboards — No waiting for IT, no developer bottlenecks, no prioritization queues
Rapid iteration — Test a new metric definition in 10 seconds, not a two-week BI sprint
Consistency across branches — Same AI, same SQL patterns, same interpretation — every query, every branch, every time
ICT is promoted, not replaced — The reporting team stops writing repetitive MIS queries and becomes the Data Governance team — owning schema, access controls, and semantic validation

Powered by Qwen3-Max & Alibaba Cloud

Our solution is built entirely on the Alibaba Cloud AI ecosystem, with Qwen3-Max as the single AI engine powering every capability end-to-end.

Qwen3-Max — The Core AI Engine

Qwen3-Max via DashScope drives every step of the DatAsk experience:

Schema-aware SQL generation — Reads the database schema, foreign keys, and column descriptions before generating each query; understands Vietnamese column names and business terms without translation
Complex joins and aggregations — Handles multi-table joins, window functions, CTEs, and nested subqueries reliably — the benchmark top-tier performance on the Spider text-to-SQL dataset translates directly to real enterprise workloads
Self-correction loop — When generated SQL fails validation or returns an error, Qwen3-Max retries with the error message as context, pushing end-to-end accuracy above 90%
Chart-type selection — Classifies every result and picks the right visualization (bar, line, pie, grouped bar, area, number) automatically based on the shape of the data
Cross-language query handling — An English-speaking regional manager and a Vietnamese branch manager can ask the same question in their own language and get the same underlying answer, with banking-aware handling of product names, regulatory terms, and Vietnamese financial jargon
Natural-language insight generation — After the chart renders, Qwen3-Max (manually triggered) writes a short plain-language interpretation, labeled clearly as AI-generated so users never confuse it with a validated metric
Prompt-to-flow compilation — Transforms the user's intent into a structured plan that can be audited, saved, and replayed with fresh data

DashScope — The Unified AI Endpoint

All Qwen3-Max calls go through DashScope with a single unified contract:

One endpoint, one model — All reasoning, classification, translation, and insight generation handled by Qwen3-Max with the same auth and the same billing
Asia-Pacific latency — Regional endpoints keep round-trip time low enough for interactive use
Predictable unit economics — ~$0.002–0.008 per query vs. ~$15–20 internal cost per manual MIS request — a 2,000× efficiency gain

Alibaba Cloud — The Complete AI Infrastructure

DatAsk runs entirely on Alibaba Cloud, providing enterprise-grade reliability and the compliance posture a bank actually needs.

Service	Role in Our Solution
DashScope	Powers Qwen3-Max for text-to-SQL, chart classification, translation, and insight generation
RDS for MySQL	Read-only data warehouse mirror — where every generated SQL query actually runs
Function Compute	Stateless SQL execution, chart rendering pre-processing, and artifact save/fetch endpoints
Object Storage Service (OSS)	Stores published dashboard snapshots, exported PDFs, and cached chart images
API Gateway	Rate-limiting, authentication, and request routing for the DatAsk web app and embedded share links
ECS	Web application servers in the Asia-Pacific region (low latency to Vietnam)
Cloud Monitor	Tracks SQL error rate, Qwen retry rate, and end-to-end accuracy — feeds the governance dashboard

Why Alibaba Cloud?

Unified AI + data ecosystem — Qwen models, RDS, Function Compute, and OSS all live in one VPC with native IAM — no cross-cloud glue, no egress surprises
Southeast Asia presence — Bangkok and Singapore regions keep latency under 50ms for Vietnamese users
Enterprise compliance — Data residency, encryption at rest and in transit, audit logs, and role-based access meet banking security requirements out of the box
Cost efficiency — Pay-per-use AI inference plus auto-scaling Function Compute means we only pay for actual query volume — no idle BI licenses

Challenges We Faced

Schema understanding in Vietnamese

Enterprise data warehouses in Vietnam mix English and Vietnamese freely — table names in English, column names in snake_case English, but business descriptions, enumerated values, and reference data often in Vietnamese. Getting the SQL generator to reason across both was non-trivial. Qwen3-Max handles this natively without a translation step, but we still had to invest in a clean schema-description layer that feeds context to every prompt.

Governance for a black-box model

Banks cannot ship answers to executives when the reasoning is hidden. "Trust me, the AI said so" is not an audit trail. We solved this by making every intermediate artifact — the generated SQL, the raw row-level data, the chart, and the AI insight — visible and labeled. The user can audit any of them before sharing.

Avoiding stale-data risk

BI tools traditionally cache heavily, which is dangerous when the underlying data is moving. DatAsk enforces a re-run, never cache policy for published dashboards — every "Refresh with latest data" click re-executes the SQL against the source. Caching only happens at the artifact metadata level, never at the numbers level.

Interactive latency

For a conversational interface to feel natural, the full loop — parse → generate SQL → execute → pick chart → render — has to complete in under 10 seconds for most queries. We hit that by parallelizing the Qwen3-Max chart classification with the SQL execution, and by keeping Function Compute warm for the hot path.

What We Learned

Qwen3-Max is production-grade for text-to-SQL. Schema-aware prompts plus the self-correction loop get us above 90% accuracy on real enterprise queries — high enough to be useful, with the governance layer catching the rest.
Qwen3-Max's speed changes the UX. Fast chart classification lets us render results before the user has finished reading the answer.
The Alibaba Cloud ecosystem is genuinely cohesive. DashScope, RDS, Function Compute, OSS, and API Gateway fit together without custom glue — we shipped the full stack in weeks, not quarters.
Governance is the wedge, not the blocker. Banks don't buy AI BI because it's magical; they buy it because every SQL query is visible, every number is auditable, and no insight is ever auto-published.
Self-service adoption is a human problem, not a tech problem. The moment we showed branch managers that they could ask a question in Vietnamese and get a chart back, we had to stop the demo — they wanted to start using it.

What's Next

Semantic Layer & Metric Governance

Today each SQL query is generated fresh from the schema. Next phase: a curated metric layer where business-critical definitions (e.g., "active customer", "NPL ratio", "month-over-month growth") are pre-defined and always reused — so "revenue" means the same thing in every department.

Metric catalog — Business owners define a metric once; Qwen3-Max references it instead of regenerating
Row-level security — Integrate with the bank's existing access model so branch managers only see their branch data
Anomaly alerts — DatAsk monitors saved metrics and pings owners when numbers move outside expected bounds

Embedded Analytics & Slack / Teams Integration

Bring DatAsk to where the work happens:

Slack / MS Teams bot — Ask "@datask what's today's transaction volume?" in any channel; get the chart inline
Embedded widgets — Drop a DatAsk artifact into Notion, Confluence, or internal portals with a one-line embed
Scheduled deliveries — "Send me yesterday's branch performance every morning at 7am" — delivered to email, Teams, or LINE

Proactive Insights & Forecasting

Move from reactive (user asks a question) to proactive (DatAsk surfaces things to pay attention to):

Weekly briefings — Auto-generated Monday-morning summary of the week's key movements per business unit, powered by Qwen3-Max over the saved metric layer
Qwen3-Max forecasting — Short-horizon forecasts for transactions, deposits, and loan volumes, with confidence intervals
Drill-down chains — When a metric drops, DatAsk pre-computes the likely contributing factors so the investigation starts from an answer, not a blank canvas

This transforms DatAsk from a self-service query tool into an always-on AI analyst embedded in the enterprise.

Tech Stack

Component	Technology
Text-to-SQL, Reasoning, Classification & Translation	Qwen3-Max (DashScope)
Data Warehouse	Alibaba Cloud RDS for MySQL (read-only mirror)
Compute	Alibaba Cloud Function Compute + ECS
Storage	Alibaba Cloud OSS (dashboards, exports, cached charts)
API Gateway & Auth	Alibaba Cloud API Gateway
Observability	Alibaba Cloud Monitor + custom governance dashboard
Frontend	React + TypeScript
Visualization	Recharts + custom Canvas layout engine
Open-source Libraries	Ant Design, React Flow, Axios, TanStack Query

The Vision

"Every decision-maker deserves data they can trust, in the language they speak, in the time they have."

Today, a manager emails ICT at 8am and makes a decision without the answer at 10am. That gap — between question and insight — is where opportunities are missed, risks go unseen, and gut feel replaces evidence.

DatAsk closes that gap. With Qwen3-Max and Alibaba Cloud, every employee becomes a data-driven employee — one natural question at a time.