Marripudi Amaresh

Building Data Co‑Pilots that work!

Submitted Nov 4, 2025

Building Data Co‑Pilots is a session on how to use agentic LLMs to reliably answer complex business questions by aggregating data across many sources—warehouse/lakehouse tables, metrics/semantic layers, feature stores, logs, and knowledge bases. We will walk through the system design that turns a plain‑English question (e.g., “Why did cancellations spike in City X yesterday?”) into an executable plan: decomposing the task, selecting tools, generating and validating SQL, auto‑joining via a semantic catalog, applying statistical checks, and returning concise narratives with charts and lineage. We will also cover governance‑by‑design: policy enforcement, PII handling, cost control, observability, and human‑in‑the‑loop review.

We will demo working implementations of Swiggy data co‑pilots that support multi‑source exploration and root‑cause workflows. You’ll see the end‑to‑end flow—from prompt to plan/DAG, semantically consistent queries across heterogeneous schemas, confidence scoring and guardrails, to final decision‑ready answers—plus the playbook we used to move from prototype to production (evaluation harnesses, MCP (Model Context Protocol)‑style adapters for data tools, caching, and failure fallbacks).

Takeaways:

  1. A reference architecture and design patterns to build reliable agentic analytics co‑pilots over messy, multi‑source data—plus a production‑readiness checklist.

  2. Demo‑backed lessons on what actually works: semantic/metrics layer abstraction, MCP‑style tool adapters, guardrails and evaluation, observability, and cost governance.

Who Will Benefit:

  1. Data engineers, analytics engineers, BI developers, and data platform owners
  2. Data scientists and ML engineers exploring RAG/agentic patterns for analytics
  3. Product managers and analytics leaders driving self‑serve insights and operational decisioning

Who are we:
Meghana Negi, Senior Manager, Data Science, Swiggy
Goda Ramkumar, Vice President | Head of Data Science at Swiggy
Amaresh Marripudi, Principal Product Manager, Swiggy

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

Jumpstart better data engineering and AI futures