How to analyze an Excel or CSV file with Julius AI (and check the answers)

#ai #data #python #tutorial

Originally published at https://kuponaut.com/blog/how-to-analyze-excel-or-csv-with-julius-ai/ — current prices and plan details are kept up to date there.

Julius AI lets you upload a spreadsheet and ask questions about it in plain English. It writes and runs Python or R code to get the answer, then shows the result as text, a table or a chart. That makes it fast, but not automatically right: the written summary can disagree with the numbers underneath it. This guide covers preparing the file, asking useful questions, getting charts, checking the work and exporting, with feature names taken from the Julius documentation and prices from julius.ai on 16 October 2026.

Step 1: prepare the file before you upload it

Julius accepts CSV (.csv) and Excel (.xlsx, .xls) files, along with JSON, Parquet and several other data formats. A few minutes of preparation saves credits later. Put one header row at the top, give every column a short, unique name ("order_date", "net_sales", "region"), and remove merged cells, subtotal rows and notes typed below the table. Keep one kind of value per column: dates as dates, amounts without currency symbols mixed into some cells.

Remove anything personal you do not need for the question. Names, emails, phone numbers and customer IDs are rarely required to find a monthly trend; replace them with a neutral ID or delete the columns. For a first test, use a small file whose answer you already know, such as last quarter's totals. That way you learn how Julius behaves before trusting it with something important.

If your workbook has several tabs, you can upload it as it is. The Julius FAQ says you can refer to individual tabs by name in your prompts, so note the exact tab names first.

Step 2: upload the file or connect a Google Sheet

In a chat, click the + button (some help pages call it the paper clip) or drag the file into the conversation window. The same menu can import from Google Drive, OneDrive, SharePoint or Google Sheets. To connect a Google Sheet, the docs ask you to share it as "Anyone with the link" with Viewer access and then paste the URL in Julius. Be aware that this setting lets anyone who has the link open the sheet, so only use it for data you are comfortable sharing that way.

Julius does not enforce a row or file-size limit, according to its FAQ; the practical limit is the memory of the cloud sandbox where the code runs. The pricing page lists 16 GB of RAM on Plus and 32 GB on Pro. For very large data, the FAQ suggests connecting a data warehouse such as BigQuery instead of uploading a file.

Step 3: clean first, then ask specific questions

Start by asking Julius to describe the data: "How many rows and columns does this file have, what are the column types, and which columns have missing values?" Write the row count down; you will use it to check later answers. The Julius data-cleaning guide suggests either a one-step prompt such as "Clean this dataset" or separate commands to standardize formats, handle missing values and remove duplicates. Separate commands take a little longer but show you exactly what was changed and how many rows were dropped.

Then ask questions the way the Julius prompting guide recommends: clear, specific, with context and the output format you want. "Analyze my sales" is vague. "Using the Orders tab, show total net_sales by month for 2025 as a table, sorted by month, and say which month was highest" is something Julius can answer and you can check. Break larger tasks into steps and refine the prompt if the first answer misses the point.

Step 4: ask for the chart you need

Charts are made with Python libraries such as Matplotlib, Seaborn and Plotly. The documentation lists line, bar, scatter, histogram, box plot, pie, heatmap and violin plots, plus more advanced types such as treemaps and Sankey diagrams. Describe the chart as precisely as the table: "Line chart of monthly net_sales for 2025, months on the x-axis in order, values in USD, label the highest point."

Look at the chart before reading the summary underneath it. If the axis starts above zero, a category is missing or months are out of order, ask Julius to fix that specific point rather than regenerating everything.

Step 5: check the work before you use the numbers

Julius shows its code in the chat as "Generating code" and then "Generated code", and you can read and copy that code and see its output inline. You do not need to understand every line. Look for the filters it applied (a year, a region, rows with missing values dropped), the column it summed and whether it used all rows. The Data Explorer, opened from the right menu or with "Expand" on a table in the chat, lets you page through the dataframes Julius created, 49 rows at a time.

Then check a few facts against the source file yourself: the row count matches what you wrote down, the highest and lowest values match a sort in Excel, and any percentage change uses the same starting and ending rows you would use. The written explanation deserves the most suspicion. Treat a correlation as a lead to investigate, not as proof of cause.

Step 6: export results and tidy up

Ask Julius to export a result table as CSV or Excel, and a download link appears in the chat. To save a chart, right-click it to copy or download it. If a download does not work, the troubleshooting page suggests checking your files, refreshing the browser or asking Julius to produce the file again.

Download what you need promptly. The Files documentation says free-plan files are kept for up to 1 hour of inactivity and paid-plan files for up to 7 days of inactivity, after which uploaded and generated files are deleted. You can also remove files yourself on the My Files page by selecting them and clicking Delete. On privacy, Julius's own documentation states that customer data is not used to train models, that files are stored in the US and that you can initiate deletion of your data. If your file contains confidential or regulated data, check those pages and your own company's rules before uploading.

DEV Community