AI language models and privacy filters power anonymous reporting website
Crowdsourcing Platform Tracks Public Bribe Patterns
- Bangladeshi developers launch 'GhushSite' to crowdsource administrative bribe patterns anonymously
- Integrated AI language models automatically filter out personal identities and sensitive data
- Platform acts as a public awareness ledger without storing unverified personal metadata
Two young Bangladeshi developers have engineered an anonymous crowdsourcing platform named 'GhushSite' to map informal extortion and bribe patterns across public services. The website, hosted at www.ghush.site, went live on Friday, 21 August 2026. The platform integrates natural language processing and automated moderation filters to prevent personal harassment while logging citizen experiences.
Without needing account creation or personal identification, users can submit data regarding the designated department, nature of service, approximate location, requested financial amount, and outcome. Within four days of its launch, claims detailing BDT 49 crore in bribe demands were submitted. Earlier metrics recorded up to 23 August highlighted 458 initial submissions totaling BDT 2.45 crore across all eight administrative divisions, led by Dhaka Division with 254 entries.
Detailing the architectural vision, co-creator Saif Kabir stated, "Everyone in Bangladesh understands the existence of bribes, but what remains missing is the specific empirical data. Standing in line at a service counter, a citizen rarely knows whether the demanded sum is the typical unofficial rate or double. We engineered this open ledger to eliminate that information asymmetry." Inspired by India's 'bribes.fyi', Saif collaborated with co-developer Shahed Sharif to program the local interface in a single day.
Data ingestion follows a three-stage automated pipeline. After submission, an integrated language model systematically parses the text to scrub personal names, phone numbers, national IDs, file trackers, and abusive language before publication. Saif explained, "The moderation engine serves as a privacy firewall rather than a factual verification mechanism. To eliminate defamation risks, users can identify departments but never individuals. Due to our permanent automated censorship filters, our underlying databases do not retain any identifiable personal metadata."
The creators reiterated that the platform does not function as an official investigative entity, but rather as an open public dataset to expose systemic patterns. Saif Kabir highlighted that multiple reports citing identical unofficial costs within a specific office establish an analytical pattern that aids citizens, researchers, and policymakers in understanding institutional trends.
//DBTech/NAP/SME/EMS//
