Back with entry #3 of my Daily Voice Diary series.
I skipped a couple of days — been busy and not the most consistent — so in this post, I’m catching up and sharing what I’ve been working on over the past few days.
What I’ve Been Focused On
The main thing I’ve been working on is wrapping up a few smaller freelancing projects. These are usually less focused on custom code and more about low-code automation — building integrations where I craft a strong prompt for a voicebot and connect it to external tools using platforms like N8N or Make.
These kinds of builds are quick, clean, and efficient. They typically range from $500 to $1,000 per project and can often be wrapped up in a few hours or a day. They’re great for maintaining momentum between larger builds.
Voicebot + ActiveCampaign
One of the main projects I’ve been wrapping up is a voicebot that integrates directly with ActiveCampaign.
The client sends mass email campaigns via ActiveCampaign. Inside those emails, each recipient sees a button that links to a brochure. When someone clicks the button, ActiveCampaign tags them as having clicked.
From there, I’ve set up a Make.com webhook that watches for those tags, pulls in the lead’s data — email, property of interest, etc. — and logs it all into a Google Sheet. Then, every afternoon, the AI voicebot goes through that list and starts calling the leads.
Outlook Calendar Integration
A second task on the same project involves setting up an automation to book meetings using the Microsoft Outlook API. I’m more used to working with Google Calendar or Cal.com, so I’ve been digging into Microsoft’s documentation to learn how their system handles event creation, availability checks, and authentication.
Real Estate AI Voice Agent
I’m also working on a real estate voice agent for another client. It’s a bit more straightforward:
The client runs Facebook ads to generate property leads
Interested users fill out a form, and their data is added to a Google Sheet
The voicebot pulls that info, makes a personalized outbound call
If the person qualifies, the AI books them into a meeting with the property owner
If they don’t pick up, the AI follows up via WhatsApp
Voice Prompting Techniques (Side Project)
Aside from client work, I’ve also been putting time into a personal side project that I’m turning into a video: a breakdown of the techniques I use when prompting voice agents.
This demo features an outbound sales agent that applies the prompting techniques in real time.
Some of the core areas I’ve been exploring include:
End-of-thought detection: Prompting the AI to detect when the user is still speaking, and pause or remain silent rather than interrupting.
Interruption handling: Teaching the AI to recover from mid-sentence interruptions naturally and stay on track without losing conversational context.
Multi-shot prompting: Giving the AI layered examples to help guide its tone and response structure more dynamically throughout a call.
Persona crafting: Experimenting with giving the AI a clear character or tone (e.g., energetic sales rep, calm consultant) to create more relatable experiences.
Natural speech patterns: Implementing human-like traits such as filler words, stutters, and SSML-based pauses to make conversations sound less robotic.
If you’re interested in seeing a full breakdown of this project in video form, check out my latest video here: https://youtu.be/xYySozWlWhY?si=m0_XrzKJTb753dL_
Back with entry #3 of my Daily Voice Diary series. I skipped a couple of days — been busy and not the most consistent — so in this post, I’m catching up and sharing what I’ve been working on over the past few days. What I’ve Been Focused On The main thing I’ve been working on is wrapping up a few smaller freelancing projects. These are usually less focused on custom code and more about low-code automation — building integrations where I craft a strong prompt for a voicebot and connect it to external tools using platforms like N8N or Make. These kinds of builds are quick, clean, and efficient. They typically range from $500 to $1,000 per project and can often be wrapped up in a few hours or a day. They’re great for maintaining momentum between larger builds. Voicebot + ActiveCampaign One of the main projects I’ve been wrapping up is a voicebot that integrates directly with ActiveCampaign. The client sends mass email campaigns via ActiveCampaign. Inside those emails, each recipient sees a button that links to a brochure. When someone clicks the button, ActiveCampaign tags them as having clicked. From there, I’ve set up a Make.com webhook that watches for those tags, pulls in the lead’s data — email, property of interest, etc. — and logs it all into a Google Sheet. Then, every afternoon, the AI voicebot goes through that list and starts calling the leads. Outlook Calendar Integration A second task on the same project involves setting up an automation to book meetings using the Microsoft Outlook API. I’m more used to working with Google Calendar or Cal.com, so I’ve been digging into Microsoft’s documentation to learn how their system handles event creation, availability checks, and authentication. Real Estate AI Voice Agent I’m also working on a real estate voice agent for another client. It’s a bit more straightforward: The client runs Facebook ads to generate property leads Interested users fill out a form, and their data is added to a Google Sheet The voicebot pulls that info, makes a personalized outbound call If the person qualifies, the AI books them into a meeting with the property owner If they don’t pick up, the AI follows up via WhatsApp Voice Prompting Techniques (Side Project) Aside from client work, I’ve also been putting time into a personal side project that I’m turning into a video: a breakdown of the techniques I use when prompting voice agents. This demo features an outbound sales agent that applies the prompting techniques in real time. Some of the core areas I’ve been exploring include: End-of-thought detection: Prompting the AI to detect when the user is still speaking, and pause or remain silent rather than interrupting. Interruption handling: Teaching the AI to recover from mid-sentence interruptions naturally and stay on track without losing conversational context. Multi-shot prompting: Giving the AI layered examples to help guide its tone and response structure more dynamically throughout a call. Persona crafting: Experimenting with giving the AI a clear character or tone (e.g., energetic sales rep, calm consultant) to create more relatable experiences. Natural speech patterns: Implementing human-like traits such as filler words, stutters, and SSML-based pauses to make conversations sound less robotic. If you’re interested in seeing a full breakdown of this project in video form, check out my latest video here: https://youtu.be/xYySozWlWhY?si=m0_XrzKJTb753dL_