Similarity API

Hey Indie Hackers — I’m Siyana. I’m building Similarity API, a hosted fuzzy matching + deduplication service for when your data is too messy (and too big) for “quick scripts” to survive.

If you’ve ever tried to dedupe a CRM export, vendor list, signups, or company names at scale, you’ve probably hit one (or all) of these:

• pairwise comparisons explode

• Spark/Python UDF approaches get slow/expensive

• LSH/MinHash or custom pipelines are powerful but take time to implement + tune

• you end up maintaining a bunch of brittle cleaning rules forever

What Similarity API does

You send a list of strings, and it returns:

• duplicate groups and/or matched pairs

• optional normalization (lowercase, punctuation removal, token sorting, etc.)

• company-name cleanup presets (e.g., stripping Inc/LLC/Ltd-style suffixes)

It’s meant to be the “matching step” you plug into whatever pipeline you already have (Databricks/Spark, dbt, scripts, ETL tools, etc.).

Benchmark

I ran a public benchmark on 1,000,000 rows and got end-to-end matching in ~7 minutes (1,000x faster than TheFuzz!), with comparisons to the common approaches people try first.

Benchmark link: https://similarity-api.com/blog/speed-benchmarks

(If you’re doing entity resolution / dedupe at scale, that page shows exactly what comes back and what the workflow looks like.)

Who it’s for

• RevOps / CRM admins cleaning duplicates

• data teams doing entity resolution

• anyone building enrichment pipelines where duplicates wreck downstream workflows

What I’m looking for

I’d love feedback from anyone who has fought duplicates in the real world:

• What data source causes you the most pain (Salesforce, HubSpot, spreadsheets, internal DBs)?

• Do you prefer “auto-merge” or “review + approve” workflows?

• What would make you trust an automated dedupe system?

If you’ve got a messy dataset story, I’m all ears — and happy to share what I’ve learned building + benchmarking this.