Cleaning your contact database in 15 minutes

April 29, 2025

Contact database maintenance the LLM way

A client had a contact database running into tens of thousands of addresses. These are the street addresses of their installed base.

Over the last many years addresses have got corrupted- Delhi became Dli or Dehli. Or something. Most Indian states and cities were mis-spelt. Worse, there was no consistency.

Sometimes it was more insidious- the city Nagpur was spelt as Nagaur and only when you looked at the state, you realised the error. Both Nagpur and Nagaur are cities but thankfully, belong to different states.

Doing this for almost a 100000 addresses is impossible; doing it accurately and without spending a lot of money or time is what we were looking at.

We ran the contact database against a mini model of Chat GPT.

One hour of coding and testing; 25 minutes for the automated script to run. And while we did not bill the customer, the cost of the api was less than a cup of coffee.

All contact database address records are clean and consistently spelt. Good fun job over the weekend.


Leave a Reply