Moving Agentic Apps from Prototype to Production Responsibly

Agentic applications are moving fast—from demos that impress in minutes to real-world systems expected to operate reliably, securely, and at scale.

But the leap from prototype to production exposes hard realities: unpredictable agent behavior, tool and integration failures, prompt drift, latency and cost spikes, and the uncomfortable truth that “it worked once” isn’t a strategy. At the same time, increased autonomy brings increased responsibility—systems must be safe, auditable, and aligned with human and business intent. In this talk, we present a practical, end-to-end blueprint for productionizing agentic apps responsibly.

We’ll cover proven architecture patterns; guardrails and safety controls; evaluation, testing, and red-teaming strategies; observability and debugging; reliable tool execution; and cost/latency optimization. We’ll also explore governance models and human-in-the-loop designs that balance autonomy with accountability, helping teams design for failure without slowing innovation.

Whether you’re building internal copilots, autonomous workflows, or customer-facing agentic products, this session will equip you to ship agents that are measurable, debuggable, and trustworthy—agents that don’t just think, but deliver, responsibly.

Evaluation Lessons for Agents in Production

Agentic applications are everywhere— from autonomous workflows to multi-step AI systems that look production-ready in demos. Once these agents hit real users, real data, and real scale, teams discover an uncomfortable truth: most agentic apps fail silently in production. When something goes wrong, it’s hard to know why, where, or how often.
We will explore a gap between prototype and production for agentic systems: lack of observability and evaluation. This talk breaks down how teams can move beyond “it worked once” to systems that are measurable, debuggable, and reliable.
We’ll walk through architectural patterns for tracing agent decisions, evaluating agent behavior over time, detecting drift and failures, and correlating cost, latency, and security at task level. You’ll see how production-grade observability and evaluation transform agentic apps—from opaque black boxes into systems you can confidently operate, scale, and improve.
This session is not about adding more agents or smarter prompts—it’s about knowing whether your agentic app is actually ready for production, and what to fix when it isn’t.

From RAG to Reliable Agents: Fixing Context Poisoning in AI-Native Workflows

Retrieval-Augmented Generation (RAG) solved the knowledge problem for enterprise AI. But as we move toward agent-driven workflows, a new class of failures is emerging—far more subtle, and far more dangerous - Context poisoning.
In AI-native workplaces, agents don’t just retrieve information—they accumulate, transform, and act on context across documents, APIs, chats, and tools. Over time, this context becomes noisy, inconsistent, and sometimes adversarial, leading to silent failures: incorrect decisions, broken workflows, and loss of trust.
In this session, we explore how context poisoning manifests in real-world enterprise systems—from long-running copilots to multi-step agent orchestration—and why traditional RAG architectures are not enough.
You’ll leave with a mental model and actionable strategies to evolve from static RAG pipelines to reliable, production-ready graph-based agent systems—especially in AI-powered workplace environments like copilots and digital assistants.
Because in the shift from retrieval to reasoning, managing context—not just generating text—is what defines whether your AI system works… or quietly fails.

India Builds in the Era of GenAI and Agentic AI

We are entering into an era where AI is just not generating content! – It’s generating capabilities. From ideas to autonomous apps, see how India is leading the AI Revolution

AWS Community Day - Technical Keynote

AI-Powered Photo Editor App

Creating image editing applications traditionally requires complex infrastructure, specialized ML expertise, and ongoing maintenance—making it costly and time-consuming for developers to build and scale. In this hands-on workshop, you'll build a complete AI-powered image editing application without managing a single server or ML model. Users can transform images with simple text prompts like "make this sunset more dramatic" or "remove the background and replace with mountains.

React India 2025
AI For Bharat Workshop

Agentic Workflows with Strands Agents and AgentCore

Learn to build AI agents for real-world apps using Strands SDK and AgentCore. Strands simplifies agent development by leveraging SOTA models to plan, chain thoughts and call tools. In this workshop, you will gain hands-on experience creating agents across diverse use cases — from fetching data, to maintaining memory, managing knowledge bases, and collaborating in multi-agent systems. We then leverage AgentCore to take your agents from POCs to production, enabling scalable & reliable deployments.

AI For Bharat Talk & Workshop

Build Generative AI powered Apps with Amazon Bedrock and Amazon Q

This open-ended workshop let participants build and explore the latest developments in Generative/Agentic AI, including content creation, applications, and conversational assistants. Leveraging AWS' generative AI, agentic AI, AI/ML, storage, database and serverless services, teams developed innovative solutions across domains. With guidance, participants conceptualized, built, and showcased their chatbot and virtual assistant creations.

Punjab AWS UG Meetup 2025