-
Build your own scraper. Brittle, and every site is different. You'll spend more time maintaining selectors than using the data.
-
Pay a heavyweight data provider. Expensive, and the data is often a stale snapshot from months ago.
-
Paste HTML into an LLM and pray. Sometimes you get valid JSON. Sometimes you get a hallucinated CEO email that doesn't exist.
I kept hitting this wall while working with lists of company domains, so I built a small API that does one thing well: send a company URL, get back clean JSON.
The two rules that shaped it
1. It reads the live site at request time. Not a database snapshot from last quarter. If a company rebranded yesterday, you get today's version.
2. It never guesses. This was the hardest constraint to enforce with an LLM in the pipeline. Missing fields come back as null — never invented. If there's no contact email on the site, you get "email": null, not a plausible-looking fake you'd import straight into your CRM.
What a call looks like
curl --request GET \
--url 'https://ai-live-company-enrichment-tech-detector.p.rapidapi.com/v1/enrich?url=https%3A%2F%2Fstripe.com' \
--header 'x-rapidapi-host: ai-live-company-enrichment-tech-detector.p.rapidapi.com' \
--header 'x-rapidapi-key: YOUR_KEY'
And the response:
{"url":"https://stripe.com","cached":false,"data":{"company_name":"Stripe, Inc.","sector":"Financial Technology / Payments","description":"Stripe is a financial infrastructure platform for businesses...","social_links":{"linkedin":"https://www.linkedin.com/company/stripe","twitter":"https://twitter.com/stripe","github":"https://github.com/stripe"},"contact_email":null,"tech_stack":["React","Next.js","Cloudflare","..."]}}
How it works under the hood
A few design decisions, for the curious:
-
Two-pass tech detection. A fast pattern-matching pass first (think Wappalyzer-style fingerprints), then an LLM enrichment pass only for what patterns can't catch. Cheaper and faster than going full-LLM on everything.
-
Hard content trimming before the LLM. Page content is capped before any model call. This keeps latency and cost predictable instead of exploding on heavy JS-rendered sites.
-
Caching with a 14-day TTL. Repeat lookups on the same domain return in ~200 ms instead of re-scraping. The
cached field in the response tells you which path you hit.
-
Strict schema validation. Every response is validated against a strict schema (Pydantic v2) before it leaves the API. Either the JSON conforms, or you get a proper error — never half-broken output.
Use cases I built it for
-
Lead enrichment: turn a list of prospect domains into CRM-ready records.
-
Tech-based targeting: filter prospects by their stack ("show me companies running Shopify").
-
Data hygiene: verify and refresh company records against the live web instead of stale databases.
Try it
There's a free tier (100 requests/month), enough to test it against your own data:
👉 AI Live Company Enrichment & Tech Detector on RapidAPI
I'd genuinely love feedback from other builders — on the positioning, the pricing, and especially: what field would you want it to extract next? Drop a comment below.