Skip to content
Automatic schema recovery

Your pipeline shouldn’t break because someone renamed a column.

When a spreadsheet arrives with different column names, SchemaHealer works out where each one belongs, rebuilds the file with the names your systems expect, and shows you exactly what it changed.

Recover a file freeSee how it works
  • No account required
  • Works with any CSV export
  • Every change explained

crm_export_q3.csv

4 columns · unexpected names found

Pipeline at risk
  • Email Addressunmapped
  • Deal Amountunmapped
  • phne_numberunmapped
  • Organizationunmapped

The next import would fail on these columns

The cost of change

Two ways a renamed column can go

Column names change every time an upstream tool ships an update. The only question is whether it turns into an incident.

  • A column gets renamed

    `Deal Amount` becomes `Opportunity Amount` in the next export.

    01

  • The import fails

    The job is looking for a column that is no longer there.

    02

  • Someone gets paged

    An engineer drops what they were doing to compare two files.

    03

  • It gets patched by hand

    A quick fix lands in the code, and nobody writes it down.

    04

  • Reports go stale

    Dashboards sit on yesterday's numbers until the data is reloaded.

    05

Downtime, an interrupted engineer, and reports nobody trusts.

What you get

Built for teams that get paged

Automatic recovery is only useful if you can check its work. Every step leaves evidence behind.

  • Prevent downtime

    Changed column names are sorted out before your import job sees them, so the run that would have failed simply works.

  • Three ways to recover

    A built-in list of known names is tried first, close matching handles typos and reworded headings, and an AI model resolves anything left.

  • Match strength you can read

    Each column shows how it was matched, from a known name through to an AI recovery, rather than an unexplained score.

  • Automatic checks

    Before the file is released it is checked for missing columns, names that could not be matched, and two columns landing on the same field.

  • A report you can share

    How much was recovered, what was matched and how, what still needs review, and whether the file is safe to use.

  • A ready-made field list

    Over three hundred standard customer and sales fields, covering contacts, companies, leads, deals, activity, marketing and support.

  • Clean file, ready to use

    Download the rebuilt file with the column names your systems expect. Column order and every value are preserved exactly.

  • Keeps working without AI

    If the AI step is unavailable, recovery still completes on the other two methods and the affected columns are flagged for review.

How it works

Nine steps, one upload

A single upload runs the whole sequence. Each step only sees what the one before it could not solve, so the simplest answer always wins.

  1. You send a CSV file. The separator is detected automatically, so exports using commas, semicolons, tabs or pipes all read correctly.

The sequence stops as soon as every column is resolved. If the known names cover the whole file, the AI model is never called.

See it work

Watch a changed file get rebuilt

Pick a sample file and follow every column from the name that arrived to the field it belongs in.

03_hubspot_export.csv

A marketing platform export. Most column names are recognised outright; the rest are close enough to match.

  • Companycompany_name
    Known name
  • Emailemail
    Known name
  • Mobile Phone Numberphone_number
    Close match
  • Lifecycle Stagecustomer_lifecycle_status
    Known name
  • Deal Amountdeal_value
    Known name
  • State/Regionstate
    Close match
  • Website URLwebsite
    Close match
  • Create Dateaccount_created_at
    Close match

Verified

8 of 8 columns recovered

Known name
4 · 50%
Close match
4 · 50%
AI match
0 · 0%
Run it on your own file

These are the engine’s real results for the sample files included with the project. Upload your own file for a live run.

About

Built for the day an export changes shape

SchemaHealer is a safety layer for the files your business runs on. It keeps a renamed column from turning into a broken report.

  • The problem

    Customer data arrives as spreadsheets exported from other systems. Every time one of those systems is updated, a column can be renamed, reworded or reordered. The import that ran fine yesterday stops working, and someone has to find out why.

  • What SchemaHealer does

    It reads the file before your systems do, works out which of your own fields each column belongs to, and rebuilds the file with the names you expect. Familiar column names are matched instantly, close variations are caught by comparison, and anything unusual is handed to an AI model that understands what the wording means.

  • Why it earns trust

    Nothing is renamed quietly. You get a report showing every column, what it became and how the decision was made, plus a list of anything that still needs a person to look at it. If the AI step is unavailable, the rest of the work still completes.

About the builder. I am a GenAI engineer working on practical applications of AI in data infrastructure. I built SchemaHealer to solve schema drift the way it actually shows up in production, with AI-assisted recovery that stays explainable end to end.

From the field

What changes when a renamed column stops being an incident

  • Supplier files used to break our nightly load about once a month, and it was always a column nobody had touched in a year. Now the run finishes and the report tells us what changed.

    Data platform lead

    B2B software, around 40 pipelines

  • The checking step is what sold the team. We are not asking anyone to trust an automatic rename, we are handing them a list of exactly what changed and what still needs a decision.

    Analytics engineer

    Financial services

  • Onboarding a new client's customer data used to be two days of matching columns by hand. Most of it now sorts itself out on the first upload, and we only look at the leftovers.

    Solutions architect

    Data consultancy

FAQ

Questions worth asking before you trust an automatic rename

If anything here is unclear, the report in the workspace shows the same information for your own file.

CSV files. The separator is detected automatically, so exports using commas, semicolons, tabs or pipes all read correctly. Anything that is not a CSV is turned away before processing starts.

Upload the file that broke last time.

You will get the rebuilt file, a column by column account of what changed, and a set of checks telling you whether it is safe to use.

Open the workspaceSee the steps

Help improve SchemaHealer

If something felt off, a column was matched badly, or there is a format you wish were supported, say so. Short notes are welcome, and they shape what gets built next.

No account needed. Nothing is shared beyond this form.