AI
2025
Software Engineer
AI Chatbot
An AI-powered chatbot for account queries, transaction assistance, and customer support automation integrated with backend banking APIs.
Metrics
Automated support flows
Reduced response time
API-backed answers
Problem
The work started here.
Customer support teams needed faster, more consistent handling for common banking questions and transaction-related requests.
Research
- Reviewed recurring account, transaction, and support questions to define the first automation scope.
- Mapped which answers could be automated and which needed handoff to support teams.
- Validated backend API touchpoints required for accurate account and transaction responses.
Planning
- Designed conversation paths for account questions, transaction lookups, and support escalation.
- Planned fallback behavior for ambiguous, sensitive, or unsupported requests.
- Separated NLP intent handling from banking API integration to keep the system maintainable.
Architecture Highlights
- Python services handled NLP intent classification and response orchestration.
- Backend integrations connected approved intents to banking APIs.
- Frontend surfaces presented guided responses and escalation states.
System Design
- Intent routing keeps account, transaction, and support workflows isolated.
- Fallback rules prevent the chatbot from answering when confidence or permissions are insufficient.
- Support handoff preserves conversation context for human follow-up.
Challenges
- Responses needed to be fast while relying on secure backend banking data.
- The assistant had to handle common requests without overstepping sensitive banking boundaries.
- Conversation flows needed to stay understandable for customers and support teams.
Solutions
- Used bounded intents, confidence checks, and explicit fallbacks for safer automation.
- Connected supported intents to backend APIs for reliable, current answers.
- Kept support escalation visible when automation was not the right path.
Performance
- Reduced repetitive support handling by automating common customer workflows.
- Improved perceived response time through guided, intent-based interactions.
- Kept the AI layer modular so intents could evolve without rewriting integrations.
Tech Stack
Python
NLP
Banking APIs
Spring Boot
React
Code Snippet
const response = await resolveIntent({
message,
customerId,
allowedScopes: user.permissions,
});Gallery & Animations
The final product uses subtle hover tilt, reveal choreography, and responsive image transitions to make the case study feel alive without turning the page into a heavy demo reel.
Lessons Learned
- AI support tools need clear boundaries before they need clever language.
- Useful automation depends on clean integrations as much as model behavior.
- Fallback states are a product feature in regulated support workflows.