Jachai AI

Instant scam & phish verification for Bangladesh. Hybrid AI + rules engine with strict client-side PII masking and instant incident recovery.

Comment

Jachai AI

Inspiration

The growing use of Mobile Financial Services (MFS) like bKash, Nagad, and Rocket in Bangladesh has also increased digital fraud. Fake lottery messages, job scams, impersonation, and OTP theft put everyday users at risk. We created Jachai AI to help people identify suspicious messages before losing money.

What It Does

Jachai AI is a Bangla-first, privacy-focused scam detection tool that:

  • Checks suspicious SMS, chats, and emails.
  • Identifies scam patterns using rules and AI.
  • Provides three risk verdicts: Strong Warning, Some Warning, and Cannot Tell.
  • Automatically masks sensitive information such as OTPs, PINs, and phone numbers.
  • Provides recovery guidance and scam awareness resources.
  • Supports both Bangla and English.

How We Built It

We built Jachai AI using a browser-based privacy system, JavaScript rule detection, and Groq AI. Sensitive information is masked before network requests. A deterministic safety system ensures that AI cannot reduce a warning identified by the local rules engine.

Challenges

Our main challenges included understanding Bangla and Banglish messages, accurately detecting sensitive information, maintaining text highlighting, and ensuring that AI-generated results could never weaken existing security warnings.

Accomplishments

  • Developed a privacy-focused scam detection platform.
  • Implemented a three-level risk assessment system.
  • Added local history and activity analytics.
  • Created 5 incident recovery guides and 8 scam awareness guides.
  • Introduced a bilingual, mobile-friendly interface.

What We Learned

We learned that effective digital security tools must be simple, privacy-focused, and accessible in local languages. Combining deterministic rules with AI provides a better balance between safety and usability.

What's Next

Our future plans include on-device screenshot OCR, expanding the scam detection rule database, and collaborating with legal and consumer rights organizations to improve recovery support.

Built With

Share this project:

Updates

Submission history