Case studySocial Rise Studio · Horizon Beam · Dec 2024 to Jul 2026
The software behind millions of Snapchat conversations
I built all of Social Rise Studio's technology, end to end, through Horizon Beam: bots that held about 3 million conversations on Snapchat in six months, a phone farm where AI drove every phone, and the software 150 virtual assistants worked in.
conversations on Snapchat, in six months
about 3 million
Snapchat accounts running at one time
10,000+
virtual assistants working in software I built
150
Social Rise Studioa phone farm, run as one system
conversations
about3 million
over six months
orchestratorrouting
post→ phone 07
check in→ phone 04
reply→ phone 08
new chat→ phone 01
follow up→ phone 06
reply→ phone 03
new chat→ phone 05
reply→ phone 02
follow up→ phone 07
reply→ phone 04
check in→ phone 08
post→ phone 01
01
02
03
04
05
06
07
08
Fact sheet
An influencer marketing company whose business ran on conversations. I reverse engineered Snapchat's private API, built the bots that did the talking and a phone farm run by AI, and the console that ran 10,000+ Snapchat and 5,000+ Instagram accounts at once.
Built all of its technology, end to end, and managed the virtual assistants who worked in it
When
Dec 2024 to Jul 2026
Who else
Me, through Horizon Beam, on every system. A team of virtual assistants worked in the software I built, and I managed them.
Stack
TypeScript
React 19
Vite
Express
MongoDB
Socket.IO
Browser extensions
AI
A phone farm
The problem
Social Rise Studio is an influencer marketing company, and its business ran on conversations: people talked to one by one, on Snapchat and Instagram, at a volume no team could type by hand. From December 2024 to July 2026 I built all of its technology through Horizon Beam.
Snapchat doesn't give software a public way to hold a conversation. So before anything could be automated, I had to get software talking to Snapchat at all.
And it was never only software. Thousands of accounts, a farm of phones and 150 virtual assistants had to run as one system, every day.
Constraints
Snapchat has no public API for conversations, so everything ran on its private one, which I worked out myself.
The bots had to post, reply and hold a conversation the way a person does.
Thousands of Snapchat and Instagram accounts ran at the same time.
A large team of virtual assistants worked in the same system, so each person could only see the accounts and servers assigned to them.
A server can only carry so many accounts at once.
Decisions
Chose
Give every staff login only the accounts and servers assigned to it, with a super admin above them
Over
One shared view where everyone saw everything
Because
with that many people in one system, staff could see each other's accounts and servers, and each of them should only see their own
Chose
Assign accounts to servers in bulk, by each server's capacity
Over
Placing them one at a time
Because
each server could only carry so many, and there were thousands of accounts
Chose
Build software to run the assistant team
Over
Managing the team by hand
Because
the team was large, and I was the one managing it
Snapchat: millions of conversations
The centre of it was Snapchat. I built a Snapchat extension and the bots behind it: they could post, reply and hold a conversation the way a person does.
Over six months they talked with about 3 million people. Every one of those had to answer what the other person actually said, and keep going.
Reverse engineering Snapchat's private API
Snapchat gives software no public way to do this. Its app talks to its servers over a private API with no documentation, so I worked it out myself and built everything else on top of it.
At the time it felt close to impossible, and it's the part of this work I'm proudest of. How it works stays private.
Instagram
On Instagram I built an extension, the AI API behind it and an app. Same job as on Snapchat: post, reply, hold the conversation. It held 750 conversations.
A phone farm run by AI
We ran the entire Snapchat operation, and that included a phone farm. Every phone in it was driven by AI: we told it what needed doing, and it did it.
Thousands of accounts at once
At one time 10,000+ Snapchat accounts and 5,000+ Instagram accounts were running. That many accounts is an operations problem as much as a software one.
So I built an operations console for it: every account, the server it ran on and its status, and who on the team owned it. Servers were filled by capacity, and the jobs that came up every day had bulk tools. It was one TypeScript app: React in the browser, Express and MongoDB on the server, one process serving both.
Reddit, X and LinkedIn
The same idea went to Reddit, X and LinkedIn. On Reddit it was an extension and a whole managed platform that did marketing for people: it found the leads, worked out what to post, posted it, talked with the people who answered and converted them. We built the same for X and for LinkedIn.
Software for the assistant team
Not everything was a bot. I managed 150 virtual assistants who worked alongside the software, for Social Rise Studio and companies like it.
So I built the software they worked in. A super admin saw everything; every assistant saw only the accounts and servers assigned to them, and the server enforced that on every request, not just the screen.
How it fits together
The studio decided what needed doing. The operations console held every account, server and assistant, and the assistants worked in it through their own logins.
The conversation AI did the talking through the extensions on Snapchat and Instagram. The phone farm took instructions and carried them out on its phones. The same pattern ran on Reddit, X and LinkedIn.
Drawing
The system
The studio
decides what needs doing
→ leads to Operations console
→ leads to Phone farmwhat needs doing
Virtual assistants
each sees only their own
→ leads to Operations consolescoped logins
Operations console
accounts, servers, roles
→ leads to Serversassigns by capacity
→ leads to MongoDB
Servers
filled by capacity
→ leads to Browser extensions
Conversation AI
posts, replies, talks
→ leads to Browser extensionsthe talking
→ leads to Reddit, X, LinkedIn
Browser extensions
Snapchat and Instagram
→ leads to Snapchat
→ leads to Instagram
Instagram
an AI API and an app
MongoDB
operations data
Phone farm
AI drives every phone
→ leads to Snapchat
Snapchat
through its private API
Reddit, X, LinkedIn
find, post, talk, convert
The console, the conversation AI, the extensions, the phone farm and the platforms they talked to.
Results
On Snapchat, about 3 million conversations in six months. On Instagram, 750.
10,000+ Snapchat accounts and 5,000+ Instagram accounts running at one time, on one system.
150 virtual assistants working in software I built, managed by me.
It ran from December 2024 to July 2026.
The numbers
Metric
Before
After
How measured
Snapchat conversations, over six months
·no earlier figure
about 3 million
My own figure, not audited
Instagram conversations
·no earlier figure
750
My own figure, not audited
Snapchat accounts running at one time
·no earlier figure
10,000+
My own figure, not audited
Instagram accounts running at one time
·no earlier figure
5,000+
My own figure, not audited
Virtual assistants working in the software
·no earlier figure
150
My own figure; I managed them
The figures on this page are mine, from running it. They aren't audited.
PermissionSocial Rise Studio is named, and these figures are shown, at my own call. No account, handle, person, message or screen is shown, and how any of it worked stays private.
Case studySocial Rise Studio · Horizon Beam · Dec 2024 to Jul 2026
The software behind millions of Snapchat conversations
I built all of Social Rise Studio's technology, end to end, through Horizon Beam: bots that held about 3 million conversations on Snapchat in six months, a phone farm where AI drove every phone, and the software 150 virtual assistants worked in.
Fact sheet
An influencer marketing company whose business ran on conversations. I reverse engineered Snapchat's private API, built the bots that did the talking and a phone farm run by AI, and the console that ran 10,000+ Snapchat and 5,000+ Instagram accounts at once.
The problem
Social Rise Studio is an influencer marketing company, and its business ran on conversations: people talked to one by one, on Snapchat and Instagram, at a volume no team could type by hand. From December 2024 to July 2026 I built all of its technology through Horizon Beam.
Snapchat doesn't give software a public way to hold a conversation. So before anything could be automated, I had to get software talking to Snapchat at all.
And it was never only software. Thousands of accounts, a farm of phones and 150 virtual assistants had to run as one system, every day.
Constraints
Decisions
Snapchat: millions of conversations
The centre of it was Snapchat. I built a Snapchat extension and the bots behind it: they could post, reply and hold a conversation the way a person does.
Over six months they talked with about 3 million people. Every one of those had to answer what the other person actually said, and keep going.
Reverse engineering Snapchat's private API
Snapchat gives software no public way to do this. Its app talks to its servers over a private API with no documentation, so I worked it out myself and built everything else on top of it.
At the time it felt close to impossible, and it's the part of this work I'm proudest of. How it works stays private.
Instagram
On Instagram I built an extension, the AI API behind it and an app. Same job as on Snapchat: post, reply, hold the conversation. It held 750 conversations.
A phone farm run by AI
We ran the entire Snapchat operation, and that included a phone farm. Every phone in it was driven by AI: we told it what needed doing, and it did it.
Thousands of accounts at once
At one time 10,000+ Snapchat accounts and 5,000+ Instagram accounts were running. That many accounts is an operations problem as much as a software one.
So I built an operations console for it: every account, the server it ran on and its status, and who on the team owned it. Servers were filled by capacity, and the jobs that came up every day had bulk tools. It was one TypeScript app: React in the browser, Express and MongoDB on the server, one process serving both.
Reddit, X and LinkedIn
The same idea went to Reddit, X and LinkedIn. On Reddit it was an extension and a whole managed platform that did marketing for people: it found the leads, worked out what to post, posted it, talked with the people who answered and converted them. We built the same for X and for LinkedIn.
Software for the assistant team
Not everything was a bot. I managed 150 virtual assistants who worked alongside the software, for Social Rise Studio and companies like it.
So I built the software they worked in. A super admin saw everything; every assistant saw only the accounts and servers assigned to them, and the server enforced that on every request, not just the screen.
How it fits together
The studio decided what needed doing. The operations console held every account, server and assistant, and the assistants worked in it through their own logins.
The conversation AI did the talking through the extensions on Snapchat and Instagram. The phone farm took instructions and carried them out on its phones. The same pattern ran on Reddit, X and LinkedIn.
Drawing
The system
The studio
decides what needs doing
Virtual assistants
each sees only their own
Operations console
accounts, servers, roles
Servers
filled by capacity
Conversation AI
posts, replies, talks
Browser extensions
Snapchat and Instagram
Instagram
an AI API and an app
MongoDB
operations data
Phone farm
AI drives every phone
Snapchat
through its private API
Reddit, X, LinkedIn
find, post, talk, convert
Results
The numbers
The figures on this page are mine, from running it. They aren't audited.
PermissionSocial Rise Studio is named, and these figures are shown, at my own call. No account, handle, person, message or screen is shown, and how any of it worked stays private.