When AI Stops Waiting: Building for Agentic Workloads

Join us in Mumbai for an exciting ClickHouse meetup bringing together engineers, architects, and data leaders to explore how modern teams are tackling the next frontier of AI-powered analytics. From agentic workloads and semantic data interfaces to real-time pipelines and production migrations, speakers will share hands-on lessons from building high-performance systems with ClickHouse and in the world of Agentic AI. Connect with the local data community and discover what it really takes to make AI and analytics work together at scale.

Don't miss out! RSVP and secure your spot!

🗓️ Agenda:

• 10:00 AM: Registration & networking
• 11:00 AM: Welcome & opening
• 11:10 AM: Talk 1 - Pushdown or pay up: predicate pushdown in pg_clickhouse by Indranil Chandra, Architect - ML & Data Engineer, Upstox
• 11:40 PM: Talk 2 - From Prompt to PASS/FAIL: Production LLM Certification on Langfuse + ClickHouse by Swaroop Oggu, Senior Solutions Architect @ ClickHouse
• 12:10 PM: Closing Remarks
• 12:20 PM: Lunch & networking

If anyone from the community is interested in sharing a talk at future events, complete this CFP form and we’ll be in touch.

🎤 Session Details: Pushdown or pay up: predicate pushdown in pg_clickhouse
Description: A cloud bill that still hurts is the best teacher. This talk is built around a real federation failure involving AWS Athena and Redshift, where missing predicate pushdown created a triple-cost disaster: full table scans, unfiltered S3 dumps, and a lifecycle policy no one remembered to set. From that war story, we move into a practical examination of cross-engine federation with ClickHouse, including when to use a Postgres-compatible interface like pg_clickhouse versus querying directly, how to verify pushdown behavior using EXPLAIN VERBOSE, and what the v0.2.0 HTTP streaming improvement means for memory pressure in production. You will leave with a concrete checklist for trusting any cross-engine query pattern before it trusts itself with your budget.
Speaker: Indranil Chandra, Architect - ML & Data Engineer, Upstox
I’m a creative technologist with a decade of experience designing AI and data-driven solutions that are both scalable and customer-centric. At the intersection of cloud-native architecture and GenAI, I specialize in building data products and platforms that thrive in highly regulated industries.

🎤 From Prompt to PASS/FAIL: Production LLM Certification on Langfuse + ClickHouse
Description: Every team shipping LLM features into production hits the same wall around month three: the model works in eval notebooks but nobody can prove what it's doing in production. Evals tell you the model "scored 87% on a held-out set last Tuesday." Certification gives you a PASS/FAIL artifact, item-level lineage, and a queryable audit trail that survives the engineer who built it leaving the company. Different deliverables. Different audiences. Most teams confuse the two and ship the wrong one.
This session walks through a working open-source pipeline that produces certifications, not just scores. It's built on Langfuse for tracing and prompt management, and on ClickHouse for everything analytical
Speaker: Swaroop Oggu, Senior Solutions Architect @ ClickHouse
Swaroop Oggu spends his days as a Solutions Architect at ClickHouse explaining why your analytical queries don't have to take forty seconds and helps clicking through postgres and lakehouses for optimizations. He came to ClickHouse after 5 years at Databricks — which is more time than he's spent on any single recipe, podcast, or movie franchise, three things he is otherwise religious about.