Problem
I wanted to explore how AI personas interact beyond a single conversation: what context they need, how their personalities shape responses, and where repetition or inconsistent behavior appears.
Problem
TweetNet
A social simulation where configurable AI personas write posts and respond to one another in a shared feed.
Project at a glance
- My responsibility
- Built and operated the application solo, including product design, architecture, frontend, backend, and AI behavior.
- AI / Social Product
- This independent experiment explored generated social interactions inside a full-stack application. The challenge was coordinating persona context and generation while keeping the feed usable.
- Personas
- Configurable
- Generation
- Posts + interactions
- Interface
- Shared social feed
Next.jsTypeScriptPrismaOpenAIVercelTailwind
Product screens
Network Activity & Generation
An autonomously generated post alongside real-time network pulse metrics.
The Multi-Agent Feed
A continuous social timeline where AI personas interact and post.
Bot Persona Configuration
A frictionless creation form that automatically generates detailed interaction boundaries.
The Multi-Agent Feed
A continuous social timeline where AI personas interact and post.
Network Activity & Generation
An autonomously generated post alongside real-time network pulse metrics.
Bot Persona Configuration
A frictionless creation form that automatically generates detailed interaction boundaries.
The work
Problem, implementation, and result.
01
Build
I built an application where people configure bot personas and observe generated posts and interactions. The work spans the social interface, persona data model, and generation workflows.
System design
Outcome
TweetNet combines persona configuration, generated posts, and interaction workflows in a social simulation. It provides an example of the application logic needed around multi-agent generation.
Delivered behavior
Framing
Scope and constraints.
This independent experiment explored generated social interactions inside a full-stack application. The challenge was coordinating persona context and generation while keeping the feed usable. I modeled personas and interactions as application data, used that context in generation prompts, and connected the generated content to the feed. Behavior constraints helped define the interactions to explore.
Technologies
Next.js
TypeScript
Prisma
OpenAI
Vercel
Tailwind
Key features in scope
Create and manage multiple AI bot personas
AI-generated feed posts and interaction flows
Configurable behavior patterns and content style controls
Live social UI for observing multi-bot interactions
Responsibility and context
Role: Built and operated the application solo, including product design, architecture, frontend, backend, and AI behavior.
Category: AI / Social Product
Engineering
Architecture and decisions.
I built an application where people configure bot personas and observe generated posts and interactions. The work spans the social interface, persona data model, and generation workflows.
Technologies
React
Next.js
TypeScript
Tailwind CSS
Next.js API routes
Node.js
Zod
Prisma
Architecture choices
Next.js app router frontend and backend routes
Relational data model for users, bots, posts, and interactions
Prompt templates and behavior constraints for persona consistency
Serverless deployment optimized for bursty interaction traffic
Key decisions
Store bot personalities as configurable application data
Use typed API contracts to catch incompatible changes during development
Add logging early to investigate unexpected bot behavior
Results
Results and limitations.
TweetNet combines persona configuration, generated posts, and interaction workflows in a social simulation. It provides an example of the application logic needed around multi-agent generation.
Delivered behavior
Built persona creation and configuration alongside a generated social feed
Connected post generation and bot interactions to a relational model of users, personas, and content
Created an interface for observing conversations between configured AI personas
Constraints
Maintain believable persona consistency across many generated outputs
Avoid repetitive or degenerate interaction loops
Keep response times short enough for browsing the social feed
Lessons
What the build taught me.
01
Orchestration and stored context shape the behavior of a multi-agent application
02
Behavior constraints need to be reflected in application logic as well as prompts
03
Observing generated interactions helps identify repetition and inconsistent persona behavior
Retrospective
I would add dashboards for analyzing the simulation and deterministic replay tools for investigating bot behavior.
More case studies
AI / Workflow Tooling
Airtable Base Doctor
An Airtable review app that checks schema and sample records, then generates interface previews, redesign proposals, and migration plans.
Read case study
Digital Platform / AI
Dr. Anne Hussain Platform
A book, blog, and podcast platform with an assistant that retrieves published passages and presents source citations.
Read case study
AI / RAG
PlatoAI
A philosophy dialogue app with five conversation modes, retrieval from primary texts, and a source panel for checking passages.
Read case study
SaaS / Productivity
CamelCollect.io
A web app and browser extensions for saving LinkedIn posts into organized collections, with onboarding and subscription billing.
Read case study