AI Agents Workshop

What You'll Learn

Six competencies. Every one of them directly employable.

🧠

Agent Architecture

Decompose any business problem into a perception-reasoning-action loop. Understand when agents are the right tool — and when a simpler approach would win.

🔁

Core Design Patterns

Master ReAct, Reflection, and Plan-Execute — the three patterns covering 80% of real-world agent problems. Learn each as a design decision, not a recipe.

🛠️

Tool Design & MCP

Design tools that teach the model how to use them correctly. Error messages that guide retries. Build and consume MCP servers shared across agents.

🕸️

Multi-Agent Orchestration

Build Orchestrator-Worker, Supervisor, Parallel Fan-Out, and Debate topologies. Know when each is right and how agents hand off context.

🛡️

Self-Healing Agents

Diagnose and recover from the 8 canonical failure modes. Implement circuit breakers, retry loops, escalation patterns, and prompt injection defences.

🚀

Production Deployment

Deploy to AWS and Azure with real security. Build LangSmith observability, LLM-as-judge evaluation suites, and a portfolio that gets you hired.

✨ The Golden Thread

One platform. Five agents. Ten layers of sophistication.

Every module builds on a single enterprise system — the Incident Intelligence Platform. Five specialist agents form a pipeline: 🔔 Alert Triage → 🔍 Investigation → 🧠 Root Cause → 🛠️ Resolution → 📋 Post-Mortem. It starts as a single triage agent on Day 1 morning. By Day 2 evening, it is a deployed, self-healing, multi-agent system with episodic memory, constitutional safety rules, an eval suite, and a live webhook trigger. Every engineer has lived through incidents — now you build the system that handles them.

The Curriculum

10 modules across 2 days. Concept → Demo → Lab in every session.

Day 1 — Design & Build

10:00 AM

🧩 Agent Blueprint

Turn a business problem into an agent architecture. The agent loop, LangGraph fundamentals, LangSmith tracing from minute one. Lab: Build the Alert Triage Agent that classifies incoming incidents with confidence thresholds.

11:30 AM

🔁 Core Patterns

Deep-dive into ReAct, Reflection, and Plan-Execute — with failure modes for each. Lab: Build the Investigation Agent using ReAct to gather evidence from logs, metrics, and alerts.

2:00 PM

💬 Prompt Engineering for Agents

Structured output, confidence calibration, 3-layer validation, retry loops that feed errors back as instructions. Lab: Build the Root Cause Agent with structured diagnosis output and confidence-gated escalation.

3:45 PM

🛠️ Tool Design + MCP + Router

Tool contracts, instructive error responses, the Router pattern. A tool returning “error: not found” teaches the model nothing. Lab: Build the Resolution Agent's tool suite (rollback, scale, restart) and package as an MCP server.

5:30 PM

🕸️ Multi-Agent Orchestration

Six topologies. Agent handoff contracts. Shared vs isolated state. Lab: Wire all 5 agents into the full Incident Intelligence pipeline — then deliberately break it. Day 2 fixes it.

5 Core Concepts Covered on Day 1

🛠️

LangGraph Fundamentals

StateGraph, nodes, conditional edges, SubGraph, Command, Send

🔁

Agentic Design Patterns

ReAct, Reflection, Plan-Execute — when to use each and how they fail

💬

Structured Output

3-layer validation: schema, semantic, business rule. Confidence calibration

🔌

MCP & Tool Design

Tool contracts with instructive errors. Build and consume MCP servers

🕸️

Multi-Agent Topologies

Orchestrator-Worker, Supervisor, Hierarchical, Fan-Out, Debate

🔎

LangSmith Tracing

Observability from minute one. Every lab traced. Debug with real data

Day 2 — Harden, Evaluate & Ship

10:00 AM

🧠 Context & Memory

4-layer memory model, context compression, durable checkpointing, Human-in-the-Loop gates. Lab: Fix the broken pipeline from Day 1 — add episodic memory so agents learn from past incidents.

11:45 AM

🛡️ Self-Healing Agents

The 8 canonical failure modes with a recovery pattern for each. Circuit breakers, prompt injection defence, cascading failure prevention. Lab: Inject 3 failure modes into the pipeline and watch it recover autonomously.

2:00 PM

⚡ Advanced Patterns

Constitutional AI — the Resolution Agent must never auto-deploy to prod without approval. Metacognition — agents that know when they don't know. Dynamic Composition — assemble responders at runtime. Lab: Add constitutional rules and metacognitive confidence gates.

3:45 PM

📊 Evaluation & Observability

LLM-as-judge, eval suites, LangSmith traces. Lab: Write 10 test cases against the Incident Intelligence pipeline. Find a failure. Fix it. Prove the improvement with data.

5:30 PM

M10

🚀 Production + Portfolio

AWS and Azure deployment. Webhook triggers. Portfolio sprint. Lab: Deploy the full Incident Intelligence Platform, trigger a live incident, and watch all 5 agents respond in LangSmith.

5 Advanced Concepts Covered on Day 2

🧠

4-Layer Memory Model

In-context, working, episodic, semantic — agents that remember and learn

🛡️

8 Failure Modes

Hallucinated args, reasoning loops, context overflow, prompt injection, cascading failure

🏛️

Constitutional AI

Agents under explicit principles that self-audit every output before acting

🤔

Metacognition

Confidence-aware agents that escalate instead of guessing. Know what they don't know

📊

LLM-as-Judge Evals

Automated evaluation suites with golden datasets and regression tracking

⚙️

Dynamic Composition

Runtime agent assembly from a registry — the pattern for open-ended workloads

Patterns That Set You Apart

Day 2's advanced module covers what separates agent engineers from agent users.

🏛️

Constitutional AI

Agents that operate under explicit principles and self-audit every output. Essential for compliance and regulated industries.

🤔

Metacognition

Agents that know when they don't know — and hand off gracefully rather than fail silently. Below threshold: escalate, don't guess.

⚙️

Dynamic Composition

Orchestrators that assemble the right agents at runtime from a registry. The pattern for truly open-ended workloads.

Your Instructor

Built by someone who ships, not just teaches.

🧑‍💻

Sridhar Jammalamadaka

Founder, FutureProof India · AI Architect

💼 16 years in IT — former Software Architect at Oracle, M.Tech from IIIT Bangalore
🤖 Built 30+ AI agents and 50+ AI chatbots for enterprise clients across healthcare, finance & legal
🎓 Trained 1,400+ professionals in Agentic AI, RAG, and LLM Engineering
🏛️ Trainer at IIIT Bangalore Executive Education
🌎 AI Consultant serving India, Dubai & USA
📚 Curriculum aligned with Anthropic's hiring benchmark & Gulli's Agentic Design Patterns

Who Is This For

Two lab modes: Builder (write code) and Strategist (write the spec). Both produce real artefacts.

👨‍💻

Backend / Full-Stack Engineers

Ready to go beyond LLM wrappers and ship real agent systems that survive production.

🤖

AI/ML Practitioners

Already working with LLMs, want to master production agent architecture and deployment.

💼

Tech Leads & Architects

Need to design agent systems and evaluate technical proposals. The Strategist track is built for you.

🎓

Ambitious Builders

Want a portfolio that gets noticed at Anthropic, Google DeepMind, Microsoft AI.

Prerequisites

Lower bar than you think.

🐍

Python

Comfortable with functions and libraries. No ML needed.

🌐

REST APIs

Know what an API call is. Everything else covered from scratch.

💻

Laptop + Browser

Everything runs in the cloud. No complex local setup.

No prior agent experience needed. No LangGraph knowledge required. No ML background. If you've written a Python function and called an API, you're ready.

Secure Your Seat

30 seats. Once full, that's it.

Workshop Pass

₹15,000

per person · GST included

✅ 2 full days of live instruction
✅ 10 hands-on labs with LangGraph + Claude
✅ Deployed 5-agent Incident Intelligence Platform on AWS/Azure
✅ Architecture Decision Record + Eval templates
✅ Lifetime recording access
✅ Telegram community + 14-day practice
✅ Portfolio review session
✅ Certificate of completion

🔥 Only 30 seats available

Full refund available up to 48 hours before Day 1.

Frequently Asked Questions

Is this for beginners or experienced engineers?+

Both. Every lab has Builder mode (write LangGraph code) and Strategist mode (write the requirements spec). Beginners build confidence on Day 1, experts hit Constitutional AI, Metacognition, and Dynamic Composition on Day 2.

Do I need to know LangGraph before attending?+

No. We cover LangGraph from scratch in Module 1. If you know Python and have made an API call, you're ready. You'll have a working agent within 90 minutes.

How is this different from other AI agent courses?+

Three things: (1) One enterprise-grade golden-thread project — a 5-agent Incident Intelligence Platform — across all 10 modules. (2) Chaos engineering — deliberately break your agents and watch them recover. (3) Aligned with Anthropic's engineering hiring benchmark.

Will I actually deploy a real agent by the end?+

Yes. Module 10 is a deployment lab. You'll deploy the full 5-agent Incident Intelligence Platform to AWS or Azure, trigger a live incident via webhook, and watch all agents respond in LangSmith traces. You leave with a running URL, not a notebook.

What if I can't attend live?+

Lifetime recording access included. All labs have written instructions. Telegram community stays active. 14-day practice challenge keeps you on track.

What's the refund policy?+

Full refund if requested more than 48 hours before Day 1. No refunds after that, but lifetime recording access ensures full value.

Ready to Build Agents That Actually Work?

10 modules. 10 labs. One deployed system. Portfolio-ready by Day 2.

📅 25 Apr & 1 May 2026

⏰ 10:00 AM – 6:00 PM

🐍 LangGraph + Claude

🔥 30 Seats Only

📧 connect@futureproofindia.com · 💬 Telegram Community