1 of 26

AI & Cloud Conference 2024

Asia's Largest

15 - 16, November 2024

Chennai Trade Center, Chennai

2 of 26

It’s me

About Me

Vikrant Guleria

A passionate technologist building innovative solutions with JavaScript & GenAI. I contribute to open-source, speak at conferences, and love sharing knowledge with the developer community. When I'm not coding, you'll find me exploring cutting-edge tech and bringing fresh ideas to life.

Software Engineer III | Tune AI

3 of 26

Live implementation of multi-agent system

4 of 26

Agenda

  • Why Agents?
  • What are AI Agents?
  • How AI Agent works?
  • Few Use case
  • Diving into - Building a Zomato Reservation Agent with Multi-agent Architecture
    • High Level Architecture
    • Components to build agents
    • Agent implementation
  • Live Demo
  • Summary
  • User Story

5 of 26

6 of 26

What are AI Agents?

🤖 AI agents function as digital assistants that can be programmed to perform specific tasks autonomously

🎯 They can help automate repetitive processes, analyze data, and handle routine operations

💡 Create value by developing AI solutions that address real business needs and improve efficiency

7 of 26

How AI Agents work

🔍 AI agents start by gathering and processing information from their environment through sensors, data inputs, or user interactions

🧠 They use sophisticated algorithms to analyze this information and understand patterns, just like a brain processing thoughts

⚡ Based on their analysis, agents make decisions and take actions autonomously, following their programmed objectives

🤝 They can collaborate with other AI agents and human users, sharing information and coordinating actions for better results

8 of 26

9 of 26

Use cases

Use cases of LLM technology has gone from email writer to more intelligent agents that can operate autonomously.

Some of the cutting edge use cases are:�

  1. Assistants to augment organisations as a 24x7 working employee
    1. Sales Assistant
    2. Healthcare Assistant
    3. Research Assistant
    4. Software Engineer Assistant�
  2. LLM technology to process data faster and cheaper
    • Processing PDFs to extract information
    • Build knowledge graphs faster
    • Reduce efforts in writing custom logic for complex documents types

10 of 26

Building a Zomato Reservation agent

11 of 26

High Level Architecture

12 of 26

Manager Agent

USER

Research Agent

Concierge Agent

Meeting Scheduler Agent

13 of 26

Manager Agent

USER

Research Agent

Concierge Agent

Meeting Scheduler Agent

Hey can you conduct some research on TechXConf

Returns Response

14 of 26

Manager Agent

USER

Research Agent

Concierge Agent

Meeting Scheduler Agent

Can you book a meeting with vikrant at Madras Square for 8 PM

Returns Response

15 of 26

Components to build Zomato Reservation Agent

16 of 26

🟠 agents-js: JavaScript package based on OpenAI swarm framework

🟠 Zomato API: Get restaurants, slots and book table

🟠 Serper APIs: Search web functionality

🟠 Tune API: Access to AI models

🟠 Zapier API: To Schedule meeting​

17 of 26

Agent implementation

18 of 26

Create AI Agents

  • Name: Identifies the agent.
  • Model: Specifies the language model in use (e.g., "gpt-4”).
  • Instructions: Provides the system prompt that defines the agent's behavior.
  • Functions: Lists the callable functions enabling the agent to perform actions or access external features.

19 of 26

Function Calling flow

20 of 26

Manager Agent

The Manager Agent acts as an orchestrator in a multi-agent system, efficiently coordinating complex tasks across specialized AI agents. Here are its key characteristics and responsibilities:

21 of 26

Research Agent

The Research Agent leverages the Serper API to perform targeted research on a given topic, crawling retrieved URLs to analyze content and generate comprehensive, accurate responses.

22 of 26

Concierge Agent

The Concierge Agent helps users find nearby restaurants, check table availability, and make reservations. It proactively fills in missing details and ensures a smooth booking experience using dedicated tools.

23 of 26

Meeting Scheduler Agent

A Google Calendar-based scheduling agent that processes meeting requests by collecting necessary details (date, time, participants), checks availability and conflicts, creates calendar events, and sends email invites to participants.

24 of 26

LIVE DEMO

https://github.com/TuneHQ/multi-agent-demo

25 of 26

User Story

Tune AI has Saas and Enterprise users. I’ll cover the Enterprise use cases from an F1 GP organiser from Middle East.

Every year they receive 1000+ RFPs for events to be organised from their clients. The team has to respond back with detailed 200+ pager proposals containing the precise details to be for the event. This proposal is then pitched back to the client by their sales team.

Our Assistant goes through 1000s of previous events and helps the sales teams build their proposals 10x faster, more accurate and creatively. It uses all the latest things like function calling, RAGs, etc. in a beautiful chat interface.

26 of 26

THANK YOU!

Software Engineer III | Tune AI

Vikrant Guleria