Agent 01 / Analysis & decision support
Data Analysis Report Agent
Turn a non-sensitive CSV and a focused decision question into a reviewable analytical report with relevant charts.
Experimental demonstration
Ask one question of your data
Minimal setup. Add one dataset and one focused business question. Calculations are performed from the uploaded data; the model interprets the question but does not invent displayed figures.
In this guide
What it’s designed to do
A generic AI summary can answer the wrong question or obscure weak data. Agent 01 interprets one focused question, validates it against the dataset, calculates the answer in code and presents only the analysis needed for that decision.
Who it’s for
Business professionals and analysts who need a fast first view of tabular data without losing sight of quality, assumptions and limitations.
Inputs
- A non-sensitive CSV file with one unique header per column
- A clear analytical or decision question
- Known exclusions, units and metric definitions to consider during review
Expected outputs
- A direct answer or recommendation matched to the question
- Up to two decision-relevant charts with accessible data tables
- A concise methodology, material caveats and next steps
The three-step workflow
- Interpret and validate
Convert the question into a structured plan, then verify the requested measure, grouping, filters and ranking against the dataset.
- Calculate and reconcile
Perform every total, ranking, comparison, trend and chart calculation deterministically, then check the result against the question.
- Communicate for the decision
Present a simple answer for simple questions and a recommendation-first dashboard for substantial questions.
A sample report
Use the bundled sample dataset, ask the example question and compare the direct answer with the ranked chart and exact data table.
Dataset: Public-domain Superstore retail transactions.
Question: Region with max sales?
Direct answer: West has the highest sales: $102,479.79.
Supporting view: Ranked sales by region, with exact values available in the data table.Methodology & future development
The demonstration uses a controlled pipeline: interpret the question, validate its business meaning against the schema, calculate in deterministic code, reconcile the result, then adapt the presentation to the decision.
STAGE 02 / DATA MODEL
Extend from one file to connected datasets
A future extension could accept multiple datasets, define keys and relationships between them, create a governed analytical data model and answer questions across the connected model. This would require relationship validation, grain and cardinality checks, metric definitions, lineage, access controls and tests that prevent double counting.
Setup guide
No installation is required for the hosted demonstration. Start with the bundled sample before using your own non-sensitive CSV.
Requirements
- A non-sensitive CSV file no larger than 750 KB
- A focused business or operational question
- Human review before using the brief for a consequential decision
How to prepare
- Start with the sample. Select the bundled public-domain Superstore dataset and ask one specific question, such as “Region with max sales?”
- Review the answer and chart. Confirm that the response uses the measure and grouping you requested. Open the chart data table when you need exact supporting values.
- Try your own safe CSV. Use a non-sensitive file with a clear header row and ask a question that names the measure and grouping where practical.
Limitations & review
- The beta reads up to 10,000 rows and 80 columns and reports when a file exceeds that row coverage.
- Observed relationships do not establish causation.
- The model interprets the question; deterministic application code calculates and validates displayed figures.
- CSV files are processed for the request and are not retained by the website; schema information and the question are sent to OpenAI for interpretation.