CRM Onboarding Agent

Architecture of the CRM AI Agent using Strands SDK, Supervisor Agent, 4 specialist agents, Claude AI, and Amazon Bedrock.
CRM AI Agent onboarding interface
Multi-agent workflow showing Data, Automation, Validation, and Scoring agents working together.
Autonomous Agent Task Execution Console showing the AI completing the full CRM onboarding workflow automatically.
Supervisor Agent coordinating four specialized AI agents to complete the CRM onboarding workflow.
Live chatbot interaction showing AI understanding a healthcare CRM question and generating a tailored pipeline recommendation.
AI-generated CRM guidance with recommended pipeline stages and actionable next steps for a healthcare business.

🔥 The Problem I Face Every Day

As a Full Stack Web Developer and CRM Solutions Provider based in the San Francisco Bay Area, I coordinate Salesforce CRM implementations for real clients including Quantum Leap Wealth (financial services firm), Open Space STL and Drone Girlz (nonprofit organizations).

Every single client onboarding follows the same exhausting pattern. I spend 3 to 5 hours manually asking the same 5 questions, researching industry-specific pipeline stages, writing automation rules, building custom field schemas, and creating email templates from scratch.

The work is repetitive, time-consuming, and completely automatable. I knew there had to be a better way.

🤖 What I Built

The NextWave CRM Onboarding Agent is an autonomous AI system that replaces my entire manual consultation workflow.

A business owner answers 5 simple questions and receives a fully customized, production-ready Salesforce CRM implementation plan in under 2 minutes - something that previously took me hours to produce manually.

Built with AWS Strands Agents SDK as the multi-agent orchestration layer, with Amazon Bedrock running Claude Sonnet 4.6 as the AI foundation.

🌐 Live Demo: crm-agent-ui.vercel.app

⚙️ How the Multi-Agent System Works

The Supervisor Agent receives the business profile and autonomously orchestrates 4 specialist agents in sequence:

Agent	What It Does
🔵 Data Agent	Generates industry-specific Salesforce pipeline stages + custom field schemas
⚡ Automation Agent	Builds lead assignment rules, stale deal reminders + 3 AI sales email templates
✅ Validation Agent	Audits config for naming conflicts, missing fields + data structure quality
📊 Scoring Agent	Calculates CRM Readiness Score 0–100 with actionable improvement checklist

The agent automatically detects the industry from the business description and generates the appropriate pipeline:

Financial Services → Compliance Check, Agreement Signed stages
Healthcare → Patient Inquiry, Treatment Plan stages
Nonprofit → Donor Relationship, Recurring Donor stages
Retail / Consulting → Custom stages based on sales process

All agent activity is tracked through Agent Lifecycle Hooks

BeforeToolCall and AfterToolCall - giving complete real-time observability into every decision the pipeline makes.

🛠️ Key Technical Features

✅ AWS Strands @tool decorator - Claude AI decides autonomously which specialist agent to call based on context
✅ Server-Sent Events (SSE) - real-time word-by-word streaming responses in React 19 frontend
✅ Agent Lifecycle Hooks - BeforeToolUseHook and AfterToolUseHook for complete pipeline observability
✅ Multi-Agent Execution Flow UI - visual Supervisor → Data → Automation → Validation → Complete
✅ CRM Readiness Score - autonomous quality metric from 0 to 100 validated by Validation Agent
✅ Industry Auto-Detection - zero manual configuration required from the user
✅ Visual Kanban Pipeline Board - generated stages displayed as interactive cards
✅ Salesforce Schema JSON Export - production-ready configuration file for direct Salesforce import
✅ AI Sales Email Templates - 3 industry-specific outreach emails generated per business
✅ Subscription & Billing Intelligence - cancellation policy, refund guidance, and plan upgrade recommendations handled as autonomous Strands tool calls
✅ Amazon Bedrock AgentCore Compliant - enterprise-ready for AWS managed production deployment
✅ Live Deployment - React frontend on Vercel, Python Flask backend with Strands SDK

💳 Real Business Scenarios Handled

The agent handles real-world business questions beyond onboarding - all implemented as Strands @tool functions that Claude AI invokes autonomously:

User: "How do I cancel my Salesforce subscription?" Agent: Explains refund policy, cancellation steps, data export options - tool call: check_subscription()

User: "Can I upgrade my plan?" Agent: Explains all plan tiers, pricing per user, unlock features - tool call: get_upgrade_options()

These are not hardcoded responses - Claude AI detects the intent and calls the appropriate tool autonomously through the Strands SDK.

🌍 Real World Impact

"This is not a demo built for a hackathon. This is a solution built from a real problem I face with real clients every single week."

🏦 Quantum Leap Wealth - 3–5 hours of manual CRM consultation replaced by 2 minutes with this agent
🌿 Open Space STL - instant professional CRM onboarding without expensive consultant fees
✈️ Drone Girlz - nonprofit CRM guidance accessible immediately at zero cost

This agent makes enterprise-level CRM consultation accessible to small businesses and nonprofits who cannot afford traditional consulting fees - democratizing access to professional CRM implementation.

📊 Evaluation & Testing

The agent was tested against 10 real business scenarios across 5 industries:

Industry	Test Result	Readiness Score
Financial Services	✅ Passed	98/100
Healthcare	✅ Passed	96/100
Nonprofit	✅ Passed	97/100
Retail	✅ Passed	95/100
Consulting	✅ Passed	98/100

Results logged in eval_results.json in the repository.

🏗️ Technical Architecture

User / Business Owner ↓ React 19 Frontend (Vercel) https://crm-agent-ui.vercel.app ↓ Flask REST Backend Server-Sent Events + Routes ↓ AWS Strands Agents SDK Multi-Agent Coordinator + Memory + @tool Registries Agent Lifecycle Hooks: BeforeToolCall · AfterToolCall ↓ Supervisor Agent (crm_engine.py) ↓ ↓ ↓ ↓ Data Automation Validation Scoring Agent Agent Agent Agent ↓ Amazon Bedrock - Claude Sonnet 4.6 global.anthropic.claude-sonnet-4-6 AWS Bedrock AgentCore Compliant

🚀 What's Next

Connect directly to the Salesforce API - so instead of generating a plan, the agent creates the actual CRM configuration in Salesforce automatically.

Full end-to-end automation:

Business answers 5 questions
Agent generates complete CRM plan
Agent connects to Salesforce API
Agent creates pipeline, fields, automations directly
Business goes live immediately - zero manual setup

Built by **Ruby Gupta* - Full Stack Web Developer & CRM Solutions Provider* NextWave Tech Studio - dnextwave.com San Francisco Bay Area

Built With

amazon-web-services
bedrock
claude
crm
flask
javascript
python
react
server

Submitted to

Agents for Humans Hackathon

Created by

I designed and built the entire project solo - from architecture to deployment. I developed the multi-agent backend using AWS Strands Agents SDK with a Supervisor Agent orchestrating 4 specialist agents (Data, Automation, Validation, Scoring) with Agent Lifecycle Hooks for observability. I built the React 19 frontend with real-time SSE streaming, deployed it on Vercel, and integrated Amazon Bedrock Claude AI. This project was inspired by real Salesforce CRM coordination work I do daily for clients including Quantum Leap Wealth and Open Space STL - replacing 3-5 hours of manual consultation with a 2-minute automated workflow.

RUBY GUPTA

Updates

RUBY GUPTA started this project — Sep 13, 2026 06:09 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.