Shivananda Golle

• • Built domain-specific Graph RAG systems for 3GPP telecom documents, combining vector embeddings with

knowledge graphs (Neo4j, AWS Neptune) to enable structured retrieval over 36K+ entities and relations.

• • Designed schema-guided extraction and structure-aware chunking strategies for LLM-based pipelines, improving entity precision and reducing noisy relations in large-scale document intelligence workflows.

• • Evaluated multiple retrieval strategies (vector, graph, hybrid) across knowledge graph configurations, benchmarking retrieval quality and factual accuracy using automated evaluation frameworks.

• • Optimized end-to-end pipeline infrastructure on AWS (S3, Lambda, Step Functions, EC2, ECR, IAM, Bedrock, Neptune), reducing LLM inference latency from 214s to 97s and cost by approximately 80%.

• • Customized Finacle-based core banking solutions for enterprise clients, configuring business logic, transaction

workflows, and product definitions using Finacle scripting and the ABCD customization tool.

• • Built and maintained data processing pipelines using SQL and Python, improving data extraction and transformation

efficiency by approximately 30% for downstream reporting and analytics.

• • Worked across Linux-based development environments with Git for version control, supporting deployment workflows and environment configuration for banking application releases.