About the customer
Industry:Telecommunication
A telecom service provider known for driving digital transformation in the country wanted to harness the potential of GenAI to achieve its goals. Alongside their aim to expand 5G coverage and the home broadband footprint in the country, they wanted to democratize data analytics via natural language interfaces. And that’s where Cloud Kinetics stepped in.
Challenge
Democratizing data analytics to align business KPIs with technical data
The company sought to introduce automation and agentic AI capabilities to help employees retrieve insights using conversational language, thereby eliminating the need for SQL knowledge.
Some of the hurdles facing the telecom major included:
Siloed expert knowledge
Inconsistent reporting
Misalignment between technical data definitions and business KPIs
Inconsistent metadata
Lack of automated lineage tracking to assess the impact of upstream data changes
Solution
Enterprise data discovery using stateful agentic AI architecture on AWS
Cloud Kinetics’ solution was an agentic AI platform utilizing a stateful agentic architecture. The use cases included:
- Business glossary lookups: Retrieving definitions and related terms from SharePoint documentation.
- Data discovery: Identifying relevant Snowflake tables and data-marts using natural language queries.
Workloads & components
Orchestration:
LangGraph on Amazon ECS to manage the "Re-act Agent" workflow (Verify Input -> Thinking -> Action -> Answer).
Intelligence:
Qwen 3-32B (via Amazon Bedrock) to handle SQL generation and intent analysis.
Knowledge Base:
Amazon Bedrock Knowledge Base with OpenSearch Serverless and Amazon Titan Text Embeddings v2 for RAG-based retrieval.
Data Integration:
A Snowflake Managed MCP (Model Context Protocol) Server for real-time querying of the Data Warehouse.
Success Metrics
Enterprise-ready performance: Accuracy, clarity and multilingual assistance
This project helped demonstrate the potential of the solution for enterprise deployment. Its multilingual capability and well structured, user-friendly responses resulted in high accuracy and clarity levels.
Accuracy:
82.3% (surpassing the 80% target)
Clarity:
91.4% for well-structured, user-friendly responses
Multilingual capability:
High-fidelity performance in both English and Bahasa Indonesia
Context awareness:
Successfully maintained conversation history across multi-turn sessions
The infrastructure is set to be rolled out for the entire organization, with plans to implement S3 Vector to reduce long-term operational costs. The next steps include deep integration connecting to the customer’s internal portal and Microsoft Teams and adding predictive modeling and automated visualization generation to further up the ante.
