Autonomous customer support chat bot for e-com with protection against errors/hallucinations
- Guarantee by Agent Risk Score
- Adjustable permissions
- Built-in fraud protection
Problem
Operators want to give autonomy to agents, it's cheaper - $300 vs $3,000 But they are afraid that they will make a mistake For example, a 1000% refund, instead of 100%
Solution
Autonomous customer support chat bot for e-com with protection against errors/hallucinations
How it works: Operator sets the permission. Bot logs actions into two databases (one open, one hashed) to detect tampering. If bot overrides the permission - system check hashes and if they match - transaction automatically covered by agent risk score.
How we built it We built a cloud microservice architecture hosted on Render. The stack includes a Streamlit frontend , OpenAI API and AWS RDS (PostgreSQL). Security is enforced via an append-only database storing SHA-256 hashes.
Challenges we ran into
Handling API timeouts required a 10-second limit and fallback responses to prevent freezing. We had to enforce strict string parsing to prevent non-numeric inputs from crashing the backend. Ensuring synchronous dual-database insertion required building a strict rollback mechanism.
Accomplishments that we're proud of:
We achieved background transaction logging with instant dashboard updates and 100% accuracy in claim status triggering based on hash comparisons.
What we learned
The #1 sales blocker for AI chatbots is the fear of financial liability for errors. A cryptographic receipt chain combined with a certification badge removes this objection.
What's next ?
Offer it it to as many Shopify developers as we can !!!!
Built With
- amazon-rds-relational-database-service
- amazon-web-services
- openai
- python
- render
- streamlit
Log in or sign up for Devpost to join the conversation.