From amnesia to photographic memory: building memory into agentic systems

Objective

You have built a few agents, wired them into a multi-agent system. What comes next? This workshop takes a focused look at one of the most critical and often underestimated parts of an agentic system — memory. Participants will come away with a practical understanding of how memory shapes agent behavior, capabilities, and user experience.

Format: Hands-on workshop
Duration: 5 hours

Agenda

Introduction

A quick grounding in agentic system fundamentals — core components (LLM, Tools, Planning, Execution, Memory), the agent execution lifecycle (Observe, Reason, Act, Reflect), and why memory is a first-class concern rather than an afterthought.

Problems with a Stateless Agentic System

What actually breaks when an agent has no state: loss of context within a conversation, no personalisation, repetitive interactions.

Hands-on Exercise 1: Build a stateless travel assistant, run it through scenarios, and identify what’s missing.

Short-Term Memory

Two practical patterns: using workflow state to store preferences, tool outputs, and intermediate decisions; and passing conversation history in the prompt to maintain continuity.

Includes a section on managing conversation history efficiently — what to preserve, what to drop, and common strategies for keeping history useful without letting it balloon.

Memory Checkpoints

Checkpoints, why they’re needed, and the LangGraph-specific concepts required: checkpointers, state persistence, workflow resumption, and interrupt-and-resume patterns.

Hands-on Exercise 2: Extend the travel assistant to retain budget, interests, destinations, and trip duration across a session, asking intelligent follow-ups and maintaining coherent context.

Long-Term Memory

Limitations of short-term memory, when long-term memory becomes necessary, and the three types — semantic, episodic, and procedural. Architectures for how memory is created, stored, and retrieved, plus production best practices.

Hands-on Exercise 3: Extend the assistant further so it recalls preferred destinations, remembers budget ranges, recommends based on past choices, and picks up conversations across sessions.

Wrap-Up and Discussion

Real trade-offs between memory approaches, common production challenges, and Q&A.

Key Takeaways

By the end of this workshop, participants will:

• Understand the role of memory in agentic systems
• Build stateless, session-aware, and persistent agents
• Implement short-term and long-term memory patterns
• Use checkpoints effectively in workflow orchestration

Who is this workshop for

• Developers and engineers building AI-powered applications who want reliable, agent-driven workflows
• ML/AI engineers exploring agentic architectures
• Tech leads and solution architects designing scalable, multi-agent systems and automation platforms
• Platform and backend engineers integrating LLMs with real-world systems, APIs, and infrastructure
• Innovation, R&D, and product teams experimenting with AI agents, orchestration frameworks, and MCP-based extensibility

Difficulty level : Intermediate

Pre-requisites

What you’ll need

• Bring your own laptop
• Git installed and an active GitHub account for code access and collaboration
• Valid API keys provisioned in advance (e.g. Gemini, GPT). Note: API keys will not be provided during the workshop.
• Familiarity with an IDE or code editor (VS Code, PyCharm, etc.)

Come prepared with

Participants should be comfortable with:

• Working knowledge of Python
• LangGraph fundamentals (graphs, nodes, edges, and state)
• The basics of agentic systems, including how a simple multi-agent system is structured and orchestrated

The instructors will start with this talk to get the fundamentals, then join the workshop for a hands-on deep dive into memory architectures for agentic systems.

About the workshop instructors

Swetha A is a Solution Consultant at Sahaj Software, specialising in multi-agentic systems and the practical application of Generative AI in software engineering. She has strong expertise in designing AI-powered systems, building multi-agent workflows, and exploring AI-assisted software development. She is passionate about understanding how emerging AI technologies can augment human creativity, improve decision-making, and transform the way software is built.
Her research has been presented at the International Conference on Data Analytics and Management, and she has spoken at multiple DevDays events and Fifth Elephant conferences, sharing her expertise and insights on AI agents, AI-assisted software development, and emerging agentic architectures.

Mahita D is a Solution Consultant at Sahaj Software, where she explores how AI can transform the way software is built. She is particularly interested in understanding what makes agentic systems effective in practice and discovering how developers can get the most out of AI-assisted development. Through talks and workshops, she shares practical insights from her work to help teams build more effectively with AI.

How to attend this workshop

This workshop is open to:
🎟️ Fifth Elephant community members — https://hasgeek.com/fifthelephant#memberships
🎟️ Ticket holders for The Fifth Elephant annual conference — https://hasgeek.com/fifthelephant/enterprise-ai-in-production-meetup#tickets

This workshop is open to 30 participants (in-person) & hybrid access for remote attendees. Seats for in-person participants will be available on first-come-first-served basis. 🎟️

Need more info?

☎️ Call: (91) 7676332020
📧 Email: info@hasgeek.com