Agent Architecture of all 7 Agents
Lead Chat (Omni Stage Agent)
Pipeline Board created by the AI Agent
CRM Stage Agent

AI Powered Auto-CRM

This is a fully AI-powered CRM system that builds itself and runs automatically using multi-agent workflows. It was created for the Google ADK Hackathon to show how complex business tools can be automated using AI.

✨ Live Demo

Try it here: https://auto-crm-ai-49793268080.us-central1.run.app Watch the video demo: https://youtu.be/zmA_eOgJ8O4

🚀 What it does

One AI agent talks to the business owner, collects business info, and builds a full CRM pipeline with stages.
Another AI agent talks to leads and moves them through the pipeline automatically.
Business owners can upload a PDF that is used by the AI to answer questions from customers.
The pipeline is fully dynamic and updates in real time.

🧩 Tech Stack

Frontend: React + Vite
Backend: FastAPI
Agents: Google Vertex AI Reasoning Engines + Google ADK
Cloud: Google Cloud Storage + Vertex AI + RAG Corpus
Libs: google-cloud-aiplatform, google-adk, pydantic, python-dotenv, google-genai

🧰 Main Features

AI builds and runs the entire CRM system
Chat interface for business setup and lead interaction
Automatic lead qualification and movement across pipeline
Real-time Kanban board and database updates
Persistent sessions and secure state management

🔧 How it works

The CRM Agent builds the structure based on a conversation with the business owner.
A RAG corpus is created from the uploaded file.
The Omni Agent uses that data to chat with leads and move them through the funnel.
Everything is dynamic and handled by just two AI agents.

✅ Why it's unique

Most CRMs need forms and setup. This one builds itself from a conversation. One agent creates the system, the other runs it. It’s ideal for small businesses that want automation without the complexity.

🔢 Testing Instructions

Go to the live app and follow the conversation with the CRM Builder.
It will ask basic info about your business and request you to upload a PDF with more details.
Confirm your input when asked.
After that, the AI will generate the full pipeline with custom stages.
Go to the Lead Chat tab to experience the customer journey.
Chat as if you were a customer. The AI will respond and automatically save your responses into the CRM.

🔄 Next Steps

Add lead import and export features
Integrate with tools like HubSpot or Salesforce
Add analytics to track performance
Let the AI edit the pipeline on user request
Add tools like web search and code execution for more advanced needs

Made with love, AI, and late nights by @alekzan

Inspiration

Many businesses want better tools, but most software is hard to set up. I’ve seen companies stick to Excel just to avoid the headache. I wanted to show how AI can build and run these tools for them.

What it does

This is a smart CRM powered by AI. One multiagent system (of 6 Agents) interact to the business owner and builds the pipeline. Another agent follows the instructions of the pipeline and talks to leads, collects their info, and moves them through the stages. The business owner only has to worry to talk with the AI agent via chat.

How we built it

I used Google ADK and Vertex AI to create two agents. One is for setting up the business pipeline. The other handles conversations with leads. I used FastAPI for the backend and React with Vite for the frontend. Everything updates in real time.

Challenges we ran into

At first, I planned to use the built-in RAG query tool from ADK, but it didn’t support using different RAG corpora for each business. To fix this, I switched to the Vertex AI library and handled RAG directly in the backend. Learning how to use Google Cloud was also a challenge in the beginning, but once I got it working, it made everything much simpler and more powerful.

Accomplishments that we're proud of

The CRM builds itself and works by itself. A business owner just answers a few questions and uploads a PDF. The AI takes care of the rest. No forms, no setup, no need to manage anything manually. This is the future of software and I am ready to start building it.

What we learned

Google ADK makes it possible to build smart systems with multiple agents. But creating something useful for real businesses still needs planning and clean architecture. I also learned how to work with Google Cloud, manage RAG documents per user, and keep everything organized and secure.

What's next for AI Powered Auto CRM

Next, I want to let users upload contact lists, connect to CRMs like HubSpot, and get reports about how leads move through stages. I also want to let the AI edit pipelines and run Python code to give better answers and read CSV or spreadsheets.