Remote Desi

How we verify

Remote Desi does not trust the word “remote”. Every posting gets one of three verdicts, and every verdict carries the sentence from the posting that justifies it.

Live numbers

34327
postings scanned (all time)
34001
live postings
536
remote AND open to India (the default list)
1943
open to India, any arrangement
30759
region-locked
702
no restriction stated, listed now
0
awaiting classification
873
companies
708
boards crawled

1.6% of the live job postings we crawl are both remote and open to India. Last refresh: 29 Oct 2026, 10:42 UTC.

Two separate things have to be proven for a job to reach the default list: that the role is remote, and that someone located in India may take it. Of the 1,943 postings we can show as open to India, 536 are described as remote by the employer, 126 as hybrid, and 1,281 name an Indian city without ever saying the role is remote — usually office roles. Only the remote ones are listed by default; the rest are one click away under “Working arrangement”, labelled for what they are.

Separately, 491 of the 2,434 postings verified as open to India are held back, not deleted: 211 have gone 90 days without the employer changing them, and 280 are marked on-site by the employer.

The three verdicts

The three tiers

  1. Structured rules. Country fields and location strings from the employer's own job board (for example a Lever country: IN, or a Greenhouse location “Remote - India”). A match makes a posting a candidate; it still goes through the exclusion patterns, and the model re-checks it whenever it is available.
  2. Exclusion patterns. A fixed list of phrases that restrict location or work authorisation (“United States only”, “EMEA only”, “located in Canada”, time-zone windows around US zones). A hit means region-locked unless the text also explicitly names India, APAC or worldwide — then tier 3 decides.
  3. A small language model that must quote. For what is left, and as a second opinion on every tier-1 match, a model reads the title, the location fields and the full description and returns a verdict with a verbatim sentence as evidence and a confidence. It is instructed not to infer anything the text does not say; “no restriction stated” must come back as unclear, never open.

When the model has not run yet. A posting whose own location or region field names India, worldwide or APAC is shown as open on that quoted field (the job page marks it “tier 1”) and is re-checked by the model as soon as it is available. Everything else, including description-only hints and conflicting signals, waits as unclear. Nothing that hits an exclusion pattern is ever shown as open. Right now 2,434 of the 2,434 open postings are shown on tier-1 evidence and still await the model check.

Every open posting is re-checked every 72 hours or whenever its description changes.

Where the jobs come from

Only sources whose public endpoints and terms allow it, always linking to the employer's own page:

Never fetched: Naukri, LinkedIn, Indeed, Glassdoor, Internshala and Remotive — their terms forbid scraping or republishing. SmartRecruiters is disabled until its robots.txt permits crawling. We honour robots.txt, make at most one request per second per host and identify our crawler with a contact address.

Pay

Pay is shown only when the employer publishes it, verbatim and in the published currency. If it is absent we show “Not disclosed”. There is no estimation, no conversion to INR and no “typical range”.

Freshness, stale and hidden postings

The report button

Every job has a “Report: not open to India” button. 2 reports (or one verified report) move the job out of the open list into “unclear” and queue a manual re-check. We store a salted hash of the reporter's address, nothing else. We track the resulting false-positive rate and aim to keep it under 5%.

What we never do

Questions or a correction? See About and Takedown.