AI for Enterprise Data by Lakehouse Days

​Calling all lakehouse geeks! e6data is hosting a hands-on meetup in Bengaluru for folks who live and breathe data infrastructure. Our topic this time? Agents!!

​What’s on the menu?

• ​Strategies for handling large, messy schemas using semantic understanding, knowledge graph-driven relationship discovery, and SQL engine-like planning to avoid errors and hallucinations.

• ​Performance optimizations such as partition key usage, optimized CTEs, early predicate pushdowns, and efficient indexing/retrieval strategies.

• ​Transition from basic query-and-result to a conversational, context-aware workflow enriched with visualizations, summaries, and smart follow-up questions.

• ​Rationale for choosing a vector database (Milvus) over traditional methods, and its integration into the system.

• ​Benchmarks, challenges, and lessons learned from deploying large-scale, high-performance recommendation systems in e-commerce.

​Speakers:

Bharath Harish, Head of Product, e6data

Topic: How We Built Text-to-SQL AI Agents to 95%+ Accuracy for Enterprise Scale

Summary:
The talk will be on how to tackle large, messy schemas using semantic understanding without requiring perfect catalogs. Bharath will explain how knowledge graph-driven relationship discovery avoids join errors common in NL2SQL and how a SQL engine-like process reduces hallucinations through planning, data profiling, and accurate SQL generation. Bharath will also cover performance-aware techniques such as partition key usage, optimized CTEs, and early predicate pushdowns, and demonstrate how we move from simple query-and-result to a conversational, context-aware data-to-insight workflow with visualizations, summaries, and smart follow-up questions. 

Time: 10:00 - 11:15 AM IST

​Harsh Sharma, Software Engineer, Flipkart FCC

Topic: Building a recommendation system using Milvus, an open-source vector database

Summary:
The talk will cover why he chose a vector database over traditional approaches, how his team integrated Milvus into their system, and the indexing and retrieval strategies used for fast and accurate recommendations. Harsh will also share performance benchmarks, practical challenges, and lessons learned from deploying this system at scale for e-commerce use cases.

Time: 11:30 - 12:30 PM IST