Here is a boring, very common job, walked through start to finish: a supplier sends a product list as a CSV export, you need it as clean JSON for an app, plus a short release note saying what changed since the last list. Nine small steps, each done in one browser tab.
I wrote this with AI help and checked every tool myself. All of them are pages on theomnikit.com, a free tools site I build. I am only describing what each page does and says, and every one of these pages states that the work happens in your browser.
The starting file
The example export looks like most exports: 31 columns, about half of them empty, headers like Product Name and UNIT PRICE (USD), and the same product listed twice because someone had merged two sheets. Here is what I do with it, in order.
Step 1: Drop the columns nobody needs
I opened the file in the CSV Column Editor. It lists every column with a sample value, so a column called Field7 can still be recognised. I pressed Remove empty columns, unticked the rest I did not need, and put the five I kept in the order the app expects with the up and down arrows.
What I liked: it writes the CSV back with its own delimiter and proper quoting, so a product description with a comma in it stays one cell. The page also warns you if two kept columns end up with the same name.
Step 2: Fix the header names
The app wants snake_case keys. I pasted the five header names, one per line, into the Case Converter. It shows eleven cases at once and converts each line separately, so Product Name became product_name and UNIT PRICE USD became unit_price_usd. I typed those back into the column editor's rename boxes before downloading.
Step 3: Remove the duplicate rows
Next, Remove Duplicate Rows. Comparing whole rows found nothing, because one copy had a trailing space. So I compared on the SKU column only and turned on ignoring upper and lower case. The page shows the rows it will remove before you download, which is the part I actually care about. It keeps the first copy by default; there is an option to keep the last one instead.
Step 4: Convert the CSV to JSON
The clean CSV went into the CSV to JSON converter. I left Header row and Detect types on and chose Objects. The detail that saved me here: type detection turns 42 and true into a number and a boolean, but keeps values like 007 or a 19-digit ID as text. SKUs with leading zeros are common, and plenty of converters quietly eat them. The table preview let me check the columns before copying.
Step 5: Check the JSON against the schema
The app already has a JSON Schema for products, so I pasted the schema and the new data into the JSON Schema Validator. It supports drafts 2020-12, 2019-09, 7, 6 and 4. It found two errors, each with the path of the failing value: a price that was an empty string, and a missing category. Clicking an error puts the cursor on that value, and the list updates as you fix things.
Step 6: Diff the new list against last month's
Now the question the release note has to answer: what changed? I pasted last month's JSON on the left and the new one on the right in JSON Diff and turned on Ignore array order, because the supplier sorts the list differently every time. It lists each added, removed and changed value by path, and key order never counts as a difference. I pressed Copy differences.
If you would rather stay in CSV, the same site has a Compare Two CSV Files page that matches rows on a key column such as an ID and shows each changed cell before and after.
Step 7: Turn the changes into a table
Price changes read best as a table. I put them in a spreadsheet (SKU, old price, new price), copied the cells and pasted them into the Markdown Table Generator. It writes a GitHub-flavoured table with the pipes lined up, and can right-align the number columns.
Step 8: Write and preview the release note
I drafted the note in the Markdown Preview: a heading, a short list of added and removed products, and the table from step 7. It renders GitHub-flavoured Markdown as you type, so a broken table shows up before anyone else sees it, and it can download the .md file.
Step 9: Keep it short
My rule is that a release note fits on one screen. The Word Counter showed words, sentences and reading time as I trimmed. I got it to under 150 words.
What I took away
None of these steps is hard. What makes the job slow is the switching: a script for one part, a spreadsheet for another, a text editor that breaks quoted fields. Doing each step on a page built for exactly that step, in the order the data flows, keeps the whole job short.
If you have a step in this kind of job that none of these covers, tell me in the comments. That is how I decide what to build next.



