VISION Ai is an AI dashboard builder that lets analysts and managers add charts, such as delay minutes by cause, in plain language or SQL, see how each number is calculated, and get scheduled PDF reports and threshold alerts. It runs each query it writes and checks the result before the chart reaches your draft, nothing goes live until you publish, and a shared board shows each reader only the rows they may see.
Email me the brochure, VISION Ai
VISION Ai is FlightAtom's AI dashboard builder and reporting software for aviation analysts, managers and data teams. People describe the chart, table or interactive component they need; VISION Ai writes the query against the data an administrator has granted them, runs it to check every field the widget uses, and adds the result to a draft they publish. Each data widget explains how its numbers are calculated, shared boards show each reader only their permitted rows, and reports, page emails and alerts run under their owner's current access.
A team needs a dashboard for a recurring reporting question. The administrator has assigned the relevant data, and the owner needs a view that can be checked and maintained.
Choose the data available to your role and the measures the dashboard needs to explain.
Describe a chart or component, inspect the result, and adjust it with AI or visible editing controls. Undo and redo keep refinement in the same draft.
Check the figures and the view, then publish the dashboard. Keep version history and reuse published widgets in a PDF report.
A published dashboard version, with a saved editing history and widgets that can be reused in reports.
The dashboard owner checks the interpretation and figures before publication. Administrators determine the data each user can access.
VISION Ai checks every change the AI proposes before it reaches your draft. A failed check goes back to the model once with its errors; if the retry fails too, the AI's change is discarded with the reason shown.
VISION Ai screens captured on 11 October 2026 with synthetic demonstration data and neutral presentation branding, which is not a product setting. Open an image to inspect the workspace.
My board compares the last 30 days with the previous period, with targets on the load factor, on-time and delay-minutes tiles. What stands out lists findings that the browser works out from the numbers on the page. Synthetic demonstration data.
The How this is calculated panel for the KPI tiles shows the SQL the visual builder generated for the board's period and notes that the reader's access rules are added when it runs. It also offers Why? breakdowns and CSV or Excel downloads. Synthetic demonstration data.
A plain-language request adds a delay-cause bar chart to the autosaved draft, and nothing goes live until Publish dashboard. VISION Ai test-runs the query-builder spec behind the change and checks it against the person's data access before the change lands. The assistant's reply in this capture was scripted for the demonstration; the checks, the chart and the draft ran in the product's own code. Synthetic demonstration data.
A widget is built from a SQL query that joins three tables. VISION Ai rewrites the query onto the person's own grants and row rules, so the preview returns only the stations this analyst manages. Synthetic demonstration data.
An administrator's reusable access profile shares four tables, three of them with column or row limits. For DELAY_EVENT it withholds the MANAGER_EMAIL column, and its row rule shows each person only the rows where MANAGER_EMAIL is their own email. Synthetic demonstration data.
Alerts show a triggered threshold alert on delay minutes for the last 7 days and an unusual-change alert that compares each week with the previous 12 and shows the usual range. A weekly page email sits below them. Synthetic demonstration data.
Access rules, agreed definitions, traceable figures and scheduled delivery apply to every dashboard, whether the AI built it or you did.
Describe a widget to the AI, build it by hand from an assigned table and its related tables, or write a SQL query that joins the tables you need, and every change lands in one autosaved draft with undo and redo. Nothing goes live until you publish, and every published version is kept and can be reopened. VISION Ai rewrites an analyst's SQL onto only the tables, columns and rows that person may see, so the database runs a query VISION Ai generated, never the typed text.
Administrators grant access by table, column and row, and row rules can match the signed-in person, so one shared dashboard can show each station manager only their own station's rows. The same rules follow the data into people's own SQL, exports, reports, alerts and AI answers. Microsoft Entra ID sign-in through Keycloak can map Keycloak roles to VISION Ai permissions, so a role change or a removed account takes effect within minutes.
Widgets built in the visual builder explain themselves in plain words, with airline shorthand such as STD and PAX spelled out, as a formula and as the SQL they compile to, and See the rows lists the records behind a number or a clicked bar. A widget written in SQL shows the statement its author wrote. Every data widget shows when its data was fetched, and its data downloads as CSV or Excel under the reader's own access.
Alerts email you when a number crosses a threshold or falls outside its usual range for that day, week or month, and once more when it is back to normal. PDF reports, built by drag and drop or with the assistant, download or arrive on a schedule with fresh data, alongside CSV data reports and daily or weekly page emails.
What stands out lists up to six findings about the page you are viewing, such as concentration, leaders, trends, unusual values and missing data, using statistical rules in your browser with no AI model. Why? splits a moved total or count by station, route or another column you choose and shows each value's share of the change against the previous period or last year.
Business metrics such as on-time performance are defined once with their calculation, time basis, time zone and exclusions, and each dashboard records the version it uses, so a later edit never silently changes an existing number. A data catalog records what each table and column means, its unit and how tables link, and administrators can draft those definitions with AI and review them before saving.
Click a bar or cell to filter the page, set a board-wide period and compare it with the previous period or last year, add targets with green, amber and red states, project a trend, and build pivot tables with totals and percentages. Save any view under a name and share its link with everyone who can see that dashboard. Discuss a widget in comment threads that @mention colleagues who can see it, and pin notes to dates on time charts.
Administrators review published widgets, and approved widgets carry a badge that the server re-checks and removes if the data or presentation changes. People can start new boards from approved widgets or save widgets from their own board as private templates. Widget health flags anything published that is failing, empty or slow, and a usage report shows administrators who uses the workspace.
Connect PostgreSQL, MySQL, MongoDB, Snowflake with password or key-pair authentication, and REST JSON APIs, with verified TLS by default and credentials encrypted at rest with AES-256-GCM and never stored in a board. PostgreSQL and MySQL queries run in read-only transactions that are always rolled back, and REST connections send only GET requests. On Snowflake, VISION Ai sends either a single read-only query it generated itself or an administrator's own SQL inside a transaction that is always rolled back.
VISION Ai puts generative AI where dashboards take the most time: writing the query, wiring the chart and drafting the report. The application runs and checks the queries and components the AI writes, inside each person's permissions, and nothing reaches a published board until its owner publishes it.
When VISION Ai writes a query for a dashboard, it runs it and checks that every field the widget uses actually came back. Custom interactive components are rendered in a headless browser first, then run in a sandbox with no network access, cookies or credentials. A failed attempt is retried once with the errors and is otherwise discarded with the reason shown, and with a widget selected, a change that would touch the rest of the board is refused.
Explore turns a data question into a proposed query, built only from the tables and columns you can see, that you can preview, refine or save as a dataset; the model never sees rows. Ask answers from your own saved datasets under your current access, passing the model at most 50 rows each time it runs one, and cannot change the dashboard. The report assistant drafts a PDF report from your fields' descriptions for you to preview and save.
Every assistant works from your data catalog, and the builder can add administrator-approved widgets and metrics unchanged, so an approved calculation is reused rather than rewritten. Every model request requires zero-data-retention routing and fails rather than fall back to a provider that keeps data, database names and definitions reach the model as fenced data rather than instructions, and AI requests are rate-limited for each person.
Agree a representative sample and a baseline from your current process. These are proposed evaluation measures, not published performance results.
An approved sample dataset, a reporting question, and the dashboard your team uses today.
Build and refine a dashboard draft, check the figures and data permissions, publish a version, and reuse a widget in a PDF report.
Dashboard preparation time; agreement with reference calculations; reviewer corrections; report preparation time.
PINAX Ai is a ready-made operational wallboard and on-time performance analysis for the airline operating day. VISION Ai is a separate workspace where people build their own dashboards, reports and alerts from the data an administrator grants them. VISION Ai does not require PINAX Ai or any other FlightAtom product.
The dashboard builder, Explore and the report assistant send table and column names, definitions and field descriptions, not rows. Ask, which answers questions from saved datasets under the person's own access, passes the model at most 50 rows each time it runs a dataset and can run several for one question, and AI drafting of the data dictionary uses a few short sample values and is designed to skip columns that look personal. Every model request requires zero-data-retention routing and fails rather than fall back to a provider that keeps data.
VISION Ai runs each query the AI writes for a dashboard and checks that every field the widget uses came back. Custom components are rendered in a headless browser before they are accepted. A failed attempt is retried once with the errors and is otherwise discarded with the reason shown. With a widget selected, a change that would touch anything else is refused, and every change stays in the draft until someone publishes the dashboard.
Yes. Administrators grant access by table, column and row, and row rules can match the signed-in person. One shared dashboard then shows each reader only their permitted rows, and the same rules govern each person's SQL, exports and AI answers and the reports and alerts they set up. Anyone can share pages of their board, or a single widget, with colleagues who can already read the data it uses.
Yes. A widget built in the visual builder explains its calculation in plain words, as a formula and, for SQL sources, as the SQL its query compiles to, and See the rows lists the records behind a number or a clicked bar. A widget written in SQL shows the statement its author wrote. Widget data downloads as CSV or Excel under your own access.
VISION Ai connects to PostgreSQL, MySQL, MongoDB, Snowflake with password or key-pair authentication, and REST JSON APIs. Connections use verified TLS by default and credentials are encrypted at rest. PostgreSQL and MySQL queries run in read-only transactions that are always rolled back, Snowflake receives single read-only queries VISION Ai generated or an administrator's SQL in a rolled-back transaction, and REST connections send only GET requests. AI-written queries and plain-language Explore work on PostgreSQL and Snowflake.
Yes. Pick an assigned table, add related tables and choose columns, then configure a chart, table or metric, preview it and add it. You can also write a SQL query that joins the tables you need. Either way the data is saved as a reusable dataset. Visible controls and AI requests edit the same autosaved draft with undo and redo, and nothing goes live until you publish.
Yes. Anyone signed in can write one SELECT or WITH query that joins the tables they need, for a widget or a dataset on their assigned connections. VISION Ai checks it against a fixed subset of SQL, replaces every table with only the columns and rows that person may see, and sends the database a query it generated, never the typed text. Tables and columns outside their access look the same as missing ones. Administrators' own SQL runs as written.
Yes. PDF reports can use datasets or published dashboard widgets, and can be downloaded or delivered on a schedule with fresh data. Alerts email you when a number crosses a threshold or is unusual for that day, week or month, and again when it is back to normal. Delivery runs under each owner's current access.
As your fleet, routes, rules or teams change, we adapt VISION Ai to match, and you approve each change before it goes live.
A focused walkthrough, with the people and decisions that matter to your team.