Pankaj Dwivedi

Chewie: An MCP-based Agentic Framework for Cloud Operations at Confluent

Submitted Jan 10, 2026

Modern incident response often requires navigating a fragmented landscape of logs, metrics, and cloud consoles, creating significant cognitive load for engineers. To address this, we built Chewie, an internal operations copilot that unifies these disparate systems into an AI-enhanced troubleshooting experience. This session details the transition from simple LLM wrappers to a centralized Tools Hub built on the Model Context Protocol (MCP), which allows teams to expose composable, machine-readable tools that AI agents can use to navigate complex, multi-cloud production environments.

Architecture & Concepts: The session focuses on the system design patterns required to build a reliable agentic workflow for infrastructure. We will discuss the conceptual framework of a “Tools Hub” where internal telemetry and deployment logs are standardized via MCP for LLM consumption. Additionally, we will cover the governance models necessary for enterprise AI, specifically how to implement auditable execution layers and “human-in-the-loop” checkpoints to ensure AI-driven investigations remain safe and transparent in mission-critical settings.

Key Takeaways:

  • Design patterns for an “Agentic Hub” using Model Context Protocol (MCP).
  • Strategies for building secured and auditable enterprise context for LLMs.
  • Frameworks for automating cross-system investigations to reduce on-call toil.

Target Audience:

Software, Backend, and Site Reliability Engineers who manage distributed systems across multiple clouds and regions.

About the speaker:

Pankaj Dwivedi is a Senior Engineer at Confluent working at the intersection of cloud infrastructure and AI-assisted operations. He focuses on building large-scale deployment platforms and exploring how AI agents can safely automate cross-system investigations and reduce cognitive load for engineers running production systems.

LinkedIn: https://www.linkedin.com/in/pankydcoder/

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