Conversational ITSM Analytics Agent
A natural language interface allowing leadership to query enterprise incident data for trends, SLA breaches, and recurring issues using plain English.
The Problem
Stakeholders struggled to extract answers from complex dashboards, leading to dependency on data analysts for basic questions.
Why Traditional Systems Failed
Dashboard filters are often rigid and intimidating for non-technical leadership. Text-to-SQL solutions often fail on enterprise-specific schemas without heavy context injection.
AI-Driven Approach
Designed a RAG-based agentic system that combines API calls to the data warehouse with LLM reasoning. The system does not just query data; it interprets the user's intent (e.g., "Why is payment failure high?") and orchestrates calls to multiple data sources. It prioritizes explainability by showing the data lineage and logic behind every answer.
High-Level Architecture
[Architecture Diagram Placeholder]
Flow: Data Ingestion → Vector Embedding → LLM Router → Response Synthesis
*Detailed interaction diagrams available upon request during interview.
Business Value
Democratized access to operational insights, reducing ad-hoc reporting requests to the data team significantly.
Key Learnings
- ✓Explainability is non-negotiable; leaders trust the answer only if they can see the source data.
- ✓Mapping vague business terms ("high severity", "recently") to precise database queries requires a robust semantic layer.
- ✓Latency matters: streaming responses improves the perceived user experience for complex analytical queries.