The Fifth Elephant
The Fifth Elephant

The Fifth Elephant 2025 Winter Edition Call for Proposals

On AI disrupting analytics - make a submission

All submissions

Rajesh RS

@rajeshrs041081

Building Self-Discovering Analytics with Praval, an open-source Agentic AI framework

Submitted Nov 10, 2025

Type of submission: 15 mins talk

Building Self-Discovering Analytics with Praval, an open-source Agentic AI framework

Abstract

Traditional analytics pipelines are brittle. Schemas change, new products launch, and data teams are trapped in a constant cycle of rebuilding ETL/ELT. This session presents a new paradigm: agentic analytics.

We will demonstrate a production-ready manufacturing analytics system built with coordinated AI agents. This is built using the open-source Praval Agentic AI framework. These agents discover data meaning on the fly, use LLM reasoning to make complex quality decisions, and coordinate without central orchestration. We’ll show you how this message-driven system (using RabbitMQ - processing 1000+ msg/sec) adapts to new product types in real-time, completely eliminating the need for pipeline code changes.

Details

This talk is a technical deep dive into an agentic system that replaces a traditional, rigid analytics pipeline. We will move beyond simple chatbot demos to show a robust, asynchronous, multi-agent system in action.

The session will be structured as follows:

  1. The Problem: Why schema-bound ETL pipelines fail in dynamic environments (our use case will be in manufacturing) and create a “change bottleneck.”

  2. The Solution: Agentic Architecture

  3. Moving from “Extract, Transform, Load” to “Discover, Reason, Coordinate.”

  4. Introducing Praval, an open-source Python framework designed to build composable, message-driven AI agents.

  5. Live System Demo & Architecture

  6. The Stack: 7 data simulators → RabbitMQ → 5 specialized agents, viz., Data Analyzer Agent, Quality Decision Agent, Vision Inspector Agent, Sink Dispatcher Agent, Chat Agent

  7. Why This Scales: Messages > REST

  8. We’ll show why a message-first architecture (RabbitMQ) beats REST APIs for agent systems, avoiding the “coordinator bottleneck.”

  9. We’ll demonstrate how Praval’s broadcast() feature enables resilient agent chaining and distributed execution.

  10. Adapting in Real-Time: The demo will show the system seamlessly processing a new, previously unseen product type.

  11. We will analyze the agent reasoning traces to show how they adapted without any developer intervention.

Key Takeaways

Audience members will leave knowing:

  • How to build schema-agnostic agents: See how LLM reasoning allows agents to adapt to new data (e.g., processing “springs” data vs. “ballpoints” data with identical code in the agent), scaling better than ETL.

  • Why message-driven beats orchestration: Understand how RabbitMQ + Praval’s agent chaining enables high-throughput (1000+ msg/sec) processing with automatic failover.

  • A practical pattern for agentic AI: Learn how to move from simple threshold logic to sophisticated LLM-based reasoning for complex decision-making, like visual quality control.

Target Audience

  • Engineers and Data Architects building AI-native data platforms.

  • Technical Leads evaluating agentic architectures vs. traditional microservices.

  • Data teams tired of brittle ETL pipelines and looking for a more resilient paradigm.

Assumes: Familiarity with Python and a technical interest in moving beyond chatbot demos to production-grade, event-driven AI systems.

Speaker Bio

Rajesh Sampathkumar builds Praval, an open-source Python framework for composable LLM agents. He architects and leads delivery for enterprise-scale GenAI applications. He has built agentic applications for recommendations, RAG, hybrid search and led the delivery of large scale event-based AI platforms. Rajesh’s unusual experience combines 12 years of core manufacturing and manufacturing quality experience (Ford, Caterpillar) with a decade of experience in AI consulting and ML/AI SaaS products.

The Praval Agentic AI framework enables large scale AI systems that self-organize and can solve problems remarkably well, and Rajesh is now exploring how multi-agent AI can be built better for a diverse range of use cases across industry, but especially in industries that need such interventions the most.

All submissions

Comments

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

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

{{ errorMsg }}

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

Hosted by

The Fifth Elephant
The Fifth Elephant
Jumpstart better data engineering and AI futures