ShopShield BD — Verify Before You Pay
Inspiration
Online shopping has made purchasing products easier, but it has also created opportunities for misleading advertisements, contradictory seller promises, and risky advance-payment requests.
In Bangladesh, many consumers discover problems only after sending money. A seller might advertise Cash on Delivery (COD) but later demand advance payment through private messages. An advertised price might change during a conversation, or important return conditions might be missing.
Reports of e-commerce complaints in Bangladesh also highlight the importance of consumer protection. In August 2026, news reports citing government figures described 58,720 e-commerce complaints involving approximately BDT 3,869 crore in claimed amounts. These are reported complaints and claims, not a verified total of fraud losses.
Sources:
This inspired a simple question:
What if buyers could examine shopping promises before making a payment?
That question became ShopShield BD.
What It Does
ShopShield BD is an AI-powered consumer protection web application designed to help Bangladeshi online shoppers identify potentially risky claims, compare seller promises, and make more informed purchasing decisions.
The application offers three main verification workflows.
1. Ad Quick Check
Users can paste advertisement text or upload screenshots.
ShopShield analyzes public advertisement claims and highlights warning signs such as:
- Advance-payment requirements
- Unclear payment conditions
- Missing return or refund information
- Unsupported authenticity claims
- Important purchase terms that require further clarification
The system explains its findings rather than generating arbitrary scam scores.
2. Ad vs Seller Chat
Users provide both the public advertisement and their private conversation with the seller.
ShopShield extracts relevant purchasing terms and compares them to identify evidence-supported contradictions involving:
- Product price
- Cash on Delivery availability
- Advance-payment requirements
- Return or refund policies
- Other relevant seller promises
For example, if an advertisement offers COD but a seller later demands full advance payment, the application can highlight the inconsistency.
A warning is not proof that a seller is fraudulent. ShopShield distinguishes verified contradictions from missing or insufficient information.
3. Page Transparency
This workflow helps users examine available seller-page information and historical web records.
Where historical snapshots are available, users can explore changes in publicly recorded seller information or claims.
Availability depends on external archival data, so missing records are not treated as evidence of fraud.
How We Built It
ShopShield BD uses a modern web architecture combining AI-assisted information extraction with deterministic verification rules.
Frontend
- React
- TypeScript
- Vite
- Tailwind CSS
Backend
- Python
- FastAPI
- Pydantic
- SQLAlchemy
AI and Analysis
- Google Gemini API
- AI-assisted structured claim extraction
- Screenshot text extraction
- Evidence-based claim comparison
- Rule-based payment-risk checks
Testing and Deployment
- Pytest
- Playwright
- GitHub
- Vercel
The application supports both Bengali and English, making its interface more accessible to different users in Bangladesh.
Challenges We Ran Into
1. Preventing unsupported AI conclusions
A major challenge was ensuring that the AI did not label legitimate sellers as scammers without sufficient evidence.
We addressed this by requiring supporting evidence for important contradictions and separating verified findings from unsupported or missing claims.
2. Extracting information from screenshots
Online advertisements and seller conversations often appear in screenshots rather than structured documents.
Handling screenshot text and converting unstructured content into comparable purchasing terms required additional processing and validation.
3. Distinguishing uncertainty from risk
Missing information does not automatically mean fraud.
We designed the analysis to communicate uncertainty and recommend manual verification when there is insufficient evidence.
4. Building a reliable end-to-end application
Integrating the React frontend, FastAPI backend, Gemini API, screenshot processing, error handling, and deployment required careful testing.
We also had to handle API failures, timeouts, and asynchronous frontend requests without displaying outdated results.
Accomplishments That We're Proud Of
- Developed and deployed a functional AI-powered consumer protection application.
- Implemented three complementary verification workflows.
- Built evidence-based advertisement and seller-chat comparison.
- Added screenshot-based input and text extraction.
- Supported both Bengali and English.
- Avoided fabricated trust scores and unsupported fraud accusations.
- Implemented automated backend and frontend tests.
- Made the application publicly accessible through Vercel.
The most important accomplishment is turning a real consumer problem into a practical application that people can try before making purchasing decisions.
What We Learned
Building ShopShield BD taught me that an AI application needs more than a powerful model.
Reliability, explainability, uncertainty handling, and user trust are equally important.
I learned how to connect AI extraction with deterministic rules, manage API failures, design accessible bilingual interfaces, and deploy a complete application.
As a solo developer, I also gained hands-on experience with full-stack engineering, testing, and product development.
What's Next
Future improvements could include:
- More comprehensive evaluation using real-world, appropriately sourced datasets
- Better support for Bangla-English mixed advertisements
- Improved historical seller-information coverage
- More robust screenshot extraction
- Usability testing with Bangladeshi consumers
- Expanded buyer education and safer-shopping guidance
Our long-term vision is to make ShopShield BD a practical, accessible consumer safety tool for Bangladesh's growing digital commerce ecosystem.
Verify Before You Pay.
Live Application: https://shopshield-bd.vercel.app
GitHub: https://github.com/naimNB/shopshield-bd
Developed by: Md Naim Babu
Institution: Pundra University of Science & Technology, Bangladesh
Participation: Solo — YUNet AIgnite Buildathon 2026
Built With
- fastapi
- google-gemini
- playwright
- pydantic
- pytest
- python
- react
- sqlalchemy
- tailwind-css
- typescript
- vercel
- vite
