Rankistan

How the scanner works

Rankistan is described as an AI powered daily leaderboard that tracks active Pakistani developers on GitHub, and the interesting part is how it stays current. Rather than rebuilding the list once, it runs as a rolling daily scanner. The pipeline discovers developers from GitHub's location search across all of Pakistan and scans more than 40,000 profiles whose account creation years span from 2000 to the present. It then re evaluates more than 20,000 candidates that show potential score movement from fresh profile, activity, and repository signals, applies strict ranking criteria, and publishes only the developers who pass every gate. To stay within GitHub's API limits, the work is distributed across 24 hourly batches, so the full ecosystem is refreshed continuously instead of in one nightly job. During discovery, each developer is fetched with profile data, recent events up to two pages or 200 events in the last 60 days, repositories, and linked social accounts for LinkedIn. The script assigns category tags such as AI or ML, Web, and DevOps by matching bio, repos, and languages against a keyword file, and those tags flow through scoring into the public data file. The leaderboard UI uses the tags when present and only recomputes them for older rows that lack them, which keeps the pipeline reproducible and the tags consistent as the keyword list changes.

Scoring and activity filters

The score is a weighted formula whose weights and caps live in a single config file, and the same math runs in both the pipeline and the Register tab so a developer sees the same number before and after a batch. The base score adds stars capped at 250, an activity score from the last 30 days, followers capped at 500, and public repos capped at 200, then multiplies by 0.5 when the account is younger than six months. Activity uses diminishing returns per UTC calendar day and per event type: the nth event of a type in a day is worth base divided by log2 of n plus one, the daily total per type is capped, and the totals are summed across the 30 day window. Push events, issues, pull requests, and releases each have their own base and daily cap. Not every GitHub account qualifies. The activity filter requires at least 30 meaningful contributions in the last 60 days, an inactivity gap of no more than 30 days, an account age of at least 30 days, more than 3 public repos, and more than 1 follower. Watching, starring, forking, or commenting does not count toward the score or the contribution gates. This design rewards sustained, public engineering work and filters out parked or empty accounts before they reach the public board.

Frontend, map, and badges

The site is a JavaScript application with several tabs. The Leaderboard shows a ranked list from the public data file with search, filters, sort, CSV export, and pagination. The Map draws an outline of Pakistan with per place counts, groups developers by normalized profile locations using a deterministic strategy that does token aware country detection and alias based city matching, and lets you click a place to list its developers. The Badge tab generates a dynamic Shields.io rank badge for a GitHub README, with Markdown, HTML, or reStructuredText snippets. The Register tab validates a profile against the pipeline criteria and shows an estimated score breakdown of stars, activity by type, followers, and repos before a batch sync. The Evolution tab explains how the scoring formula changed over time and why rules were tightened, and the About tab documents pipeline logic, filters, and scheduling. The on demand AI summary system generates a short profile summary only when a developer card is expanded, sending sanitized metadata to a Cloudflare Worker that calls Groq with server side secrets, then caches the result by username. The badge API also runs on that Worker because the static GitHub Pages site cannot serve routes. The project is under the MIT license and recorded 138 stars at its last update on 2026-08-24.

Editorial conclusion

Rankistan is written in JavaScript, released under the MIT license, and recorded 138 stars on GitHub at its last update on 2026-08-24.