Data for Life 2026 entry · Challenge VLTN-05
Ask your data in Vietnamese.
Get answers as charts.
Semanio connects directly to your existing databases, understands natural Vietnamese questions, and returns interactive charts with explanations in seconds. No SQL, no waiting on the data team.
- 12 engines
- connected directly, plus Google Sheets
- 3–5 seconds
- from a Vietnamese question to a chart
- 26+ charts
- the AI picks the right visual form
- 0 lines of SQL
- required from business users
See it in action
Four ways Semanio answers a question
One chat surface. The AI picks the right answer shape: a short answer, a chart, a dashboard, or a scheduled report.
| Unit | This quarter | Same period LY | Change |
|---|---|---|---|
| Central branch | 18.4 bn | 21.0 bn | −12.4% |
| South-West branch | 24.1 bn | 25.6 bn | −5.9% |
| Central Highlands branch | 9.7 bn | 10.0 bn | −3.1% |
Interface mock-up; every number in these examples is simulated data.
How it works
Set it up once, the whole organisation uses it
One data-literate person builds the frame. After that, anyone can ask.
- 1
Step 1 — Connect or submit files
Bring the data together, once
Connect directly to 12 common engines (PostgreSQL, MySQL, SQL Server, BigQuery, Snowflake, ClickHouse…) or Google Sheets. Credentials are encrypted and the data stays on your servers.
No central database yet? Define a standard template once, then each period people submit Excel/CSV files that match it. Rows land in the right table, and a file that drifts from the template raises a warning instead of passing silently.
- 2
Step 2 — Semantic layer
Declare the business meaning so the AI gets context right
Pick tables, label and describe each column, declare metrics and the relationships between tables. This step decides accuracy: the clearer the description, the better the answers. Treat it like onboarding a new analyst.
- 3
Step 3 — One assistant per department
Each department sees only its own slice
Group related models into a context, for example Planning and Finance, pick the AI model that fits the budget, and set what may be asked. Role-based and row-level permissions apply right here.
- 4
Step 4 — Ask and receive
Ask in Vietnamese, pin a dashboard, schedule the send
Staff ask the way they would ask a colleague. Semanio generates governed SQL, runs it on live data, and returns a chart with an explanation in seconds. Pin it to a dashboard in one click, or schedule PDF/Excel delivery over Zalo OA, Telegram and Email.
Features
Enough for the whole organisation to self-serve
Four or five disconnected tools to produce one decent report. Semanio puts them in one place.
Natural Vietnamese conversation
Understands administrative time phrases (this quarter, year to date, same period last year) and business synonyms, with multi-turn context for follow-up questions.
Semantic layer, no made-up numbers
One canonical formula per metric across the organisation. The AI only sees metadata, joins are routed safely, and every query can be reviewed as SQL.
Template-based Excel/CSV submission
Define the template once, then drag and drop many files at a time with per-file status. Files that drift from the template are flagged on the spot, and a whole submission can be revoked.
12 engines plus Google Sheets
PostgreSQL, MySQL, SQL Server, BigQuery, Snowflake, ClickHouse, DuckDB, Redshift, Trino, Databricks, SQLite, NocoBase. Connected directly, with no copy of your data elsewhere.
Scheduled multi-channel reports
Schedule once and PDF/Excel reports arrive on time over Zalo OA, Telegram and Email. A background scanner flags statistical anomalies early.
Advanced analysis built in
Cohort, RFM, funnel, growth accounting, roll-rate, Pareto. Pick a dataset and a period and it runs, instead of hand-coding each one.
Core difference
Why chatting straight to a database usually breaks, and this does not
Most Text-to-SQL products let the LLM touch physical tables. Semanio puts an arbiter in between.
An LLM straight on your tables
- Invents metric formulas, a different definition on every ask
- Row fan-out on joins, numbers inflate unnoticed
- No control over who may ask what, or see how far
- Raw records get uploaded to a public cloud service
Semanio goes through the semantic layer
- Metrics are declared once; the AI must use the approved definition
- Joins are routed safely, blocking meaningless dimension mixes and row fan-out
- Role, row and column level permissions, with a full audit log
- The AI receives structural metadata only; records never leave your servers
| Criterion | Traditional BI | AI chatbot with file upload | Semanio |
|---|---|---|---|
| Interface | Drag and drop, or code | Chat, no direct database link | Native Vietnamese, multi-turn |
| Number reliability | High, depends on engineers | Weak, prone to hallucination | Governed by the semantic layer, auditable |
| Time to a report | Days to a week | Minutes, after uploading files | Seconds for a question |
| Security | Internal | Weak, raw records go to the cloud | The AI receives metadata only |
| Automated delivery | Internal email | None | Zalo OA, Telegram, Email |
| Cost | USD 10–70 per user per month | USD 25–30 per user per month | Per-organisation subscription, unlimited users |
Security
Designed around public-sector requirements
The data belongs to the organisation. Semanio treats that as a constraint, not an option.
Data stays inside the organisation
Single-tenant deployment. Queries run in place and the AI model receives structural metadata only: metric names, dimensions, definitions.
Layered access control
Role-based access, row-level security and column-level rules. Every officer sees exactly the scope they were granted.
Complete audit trail
Key action groups are recorded: who asked what, which SQL ran, who changed a metric definition, who downloaded a report.
No vendor lock-in
The model layer is abstracted: switch between Gemini, Claude and GPT as budget dictates, and plug in open-source models running inside your own network.
Data for Life 2026
Mapped against challenge VLTN-05
The challenge states: officers and managers manually consolidate figures from many sources (email, Excel, departments) to produce reports, which is slow and error-prone at peak time.
| What the challenge asks for | What Semanio delivers |
|---|---|
| Question and answer over data in natural Vietnamese | Vietnamese-to-SQL generation with multi-turn context |
| Automatic report consolidation | Generated dashboards and scheduled reports over Zalo OA, Telegram, Email |
| Real-time visualisation | Queries run against the live source, not a stale copy |
| From connected sources (email, Excel, departments) | 12 engines, Google Sheets, and template-based Excel/CSV submission |
NCB Team · Digital government and public services · Digital economy · Data for the community
Try Semanio on your own data
Connect a source, declare your metrics, ask the first question. One working session is enough.