Sender

Automate business lead generation and outreach with Java and GenAI
View Demo · Report Bug · Request Feature

Sendur is an AI-augmented lead generation automation platform that leverages n8n workflows and AI agents to streamline the process of identifying, reviewing, and engaging potential business leads.

How It Works

Automated Lead Discovery

Sendur uses n8n, an AI Agent, and SerpAPI to search the web, on an automated schedule, for businesses based on a given description or industry profile. We've all heard of prompting... yes, you need to build a prompt to share with your AI Agent. Be warned, AI Agents can, and do, get things wrong so it's best that you are explicit in your details. In this setup, we ask our Agent to look up the business details of small businesses around the areas above North Seattle. We want our agent to send these leads back in a nicely built JSON structure (we can use this for persistence). The Agent uses OpenAI and SerpAPI as a tool. Tools are nice-to-haves that are available for your AI Agent to use, however your Agent must have access to utilize these tools (API keys or other information you can let your Agent know about). If you haven't, check out what SerpAPI is. You see the node called Structure the Leads, well n8n allows you to manipulate any data output you receive. In this case our Agent returns some JSON, hopefully a list of leads. This Code node allows us to conduct any other parsing, validation or structure manipulation on this output data. Honestly, we just want to ensure the AI Agent is giving us some useful data, in the right structure we can use in our Springboot application. Which leads us to the last Receive Scheduled Leads node. This output is sent via a POST request to Sendur (/sendur/api/leads/receive-scheduled-leads). You might be asking how we remove data duplication, I mean we don't want the same leads showing up all the time. You could use a Memory node on the AI Agent, however this is generally used for chat history context, and unfortunately we're not building a chatbot. So, we filter. Going back, you see the first Get All leads node? Here's where we send a GET request to Sendur (/sendur/api/leads/find-all) to receive all our leads. Next, in the Blocked Leads node, we parse all the business names, and we pass this to our AI Agent. We tell our agent, "hey, you see this list of business names? If you run up on them while you're searching the web, ignore them and don't add them to the final output." I'm sure you can come up with a better prompt... Lastly, as a final filter we do something similar in the Sendur application code to ensure we're not persisting duplicates.

Lead & Email Generation

Once discovered, leads are enriched with relevant details and paired with AI-generated email drafts. Maybe you will generate some nice cash flow with automation, or maybe you have deep pockets. Either way, be aware: your AI Agent uses API keys for tools like OpenAI and SerpAPI and this can become costly! However, maybe you only run these schedules once a day, or maybe you change this to be manually triggered. Just run some calculations before you set this loose. Many tools have free tiers so check those out too, and be creative!

Automating Data Record Updates

We discussed how AI Agents can get things wrong and create random false data (hallucinate) and/or incomplete data! Well, incorrect data is definitely a possibility when we use AI to automate actions. Part of the goal is to receive correct, complete, and valuable business leads. So, we use n8n nodes like the Code node for validation and formatting, but what happens when our data is correct but unknown? Our AI Agent will still produce something. What we did was protect our data output by explicitly telling our agent what to put when it can not find what it searches for: "email": "Not available". What if our search just wasn't deep enough, but we really want to produce as many business leads as possible? One option is to run another, deeper and more intentional, search. I don't know about you, but I don't want to do that work. We can, again, use the services of n8n and our AI Agent. Here's another workflow, except this time we pull leads without emails from our datastore by calling our springboot API /sendur/api/leads/no-email-scheduler, we validate our data with the Code node, and send to our Agent. But we only want leads whose email has been found. Let's use another Code node to truncate the JSON returned from our Agent. Lastly, we can send these updated leads back to our springboot api endpoint /sendur/api/leads/update-emails to update our datastore. Next, we'll talk about how we send off the generated emails with a click of a button.

Human-in-the-Loop Review

You log into the Sendur web application to review and approve both the leads and their associated emails.

Email Dispatch & Webhook Integration

Approved emails are sent via an n8n webhook trigger, initiating the outbound communication process.

Lead Persistence & Workflow Expansion

All lead data is stored and updated in the system. As the project grows, additional automations and workflows (e.g., follow-ups, CRM sync, analytics) can be layered on top.

Deploying n8n on AWS EC2

1. Spin up an EC2 instance (t2.micro is free 🤑)

• Chose Ubuntu 22.04 or Amazon Linux 2
• Ensure to configure security groups to allow traffic on port 5678 (n8n's default port)
• Keep in mind: server(s) running the application will need inbound permission, and you may want to access n8n locally
• Create and download the SSH key pair to access your instance

2. Install nodejs on EC2

• ssh into your instance
• Linux 2: (curl -sL https://rpm.nodesource.com/setup_24.x | sudo bash -) then run (sudo yum install -y nodejs)
• Ubuntu : (sudo apt update && sudo apt install -y nodejs npm)

3. Install and run n8n

• Install: sudo npm install -g n8n
• To test your setup (which is currently insecure: no TLS/HTTPS) for development you should disable secure cookies for the meantime
• Later we will add a load balancer to make this publicly accessible for (yourself, your team, your org)
• In Terminal : export N8N_SECURE_COOKIE=false
• Run : n8n

4. Accessing n8n

• Browser : ec2-server-ip:5678
• If you're slowly migrating from local to AWS be sure to update your application's configuration settings
• You'll also want to ensure your n8n Request Nodes are updated to point to your application server or local ip

5. Security Tips

• Restrict inbound traffic to n8n server using tight security groups. In dev mode this should be your IP address
• Later, we'll enable AWS WAF (Web Application Firewall) and tight log accounting
• n8n has useful AWS Nodes like the Lambda Node - we need to use IAM roles with least privilege for AWS integrations
• Our EC2 dev instance won't run up a charge (if you used t2.micro 🤑) but shut n8n down when not in use
• When we first log in we create server credentials. Please don't use admin/admin or admin/password.

Building and Running the Project

1. Build the Project

Note: Ensure application.properties file is populated! To build the backend and frontend, run the following commands:

mvn clean install

npm run build

2. Running the Project

After building, you can run the backend Spring Boot application using:

mvn spring-boot:run

The frontend can be served running:

npm run watch

3. Testing the Project

Once the project is running, open up a browser and navigate to localhost:8082 and you should see text saying All is Good!.