---

🐦 Tweeting Agent

An Autonomous AI System for Generating Humorous & Viral Tweets using Open-Source LLMs

---

📌 Overview

Tweeting Agent is an AI-powered system designed to help users generate humorous, engaging, and potentially viral tweets on any given topic. It leverages open-source Large Language Models (LLMs) orchestrated through LangGraph, enabling an iterative generate → evaluate → optimize workflow.

Unlike single-prompt tweet generators, Tweeting Agent behaves like a creative loop, refining tweets over multiple iterations until the best possible version emerges.

---

🚀 Key Features

• 🤖 Multi-Agent Architecture using LangGraph
• 🔁 Iterative Improvement Loop (Generator → Evaluator → Optimizer)
• 🧠 Fully Open-Source LLM Stack via Ollama
• 🎯 Optimized for humor, virality, clarity, and engagement
• ⚡ Modular, extensible, and framework-agnostic design
• 🧪 Easy experimentation with different models, prompts, and scoring logic

---

🧠 System Architecture

The Tweeting Agent uses three specialized LLM roles, each optimized for a specific task:

1. Generator Agent

• Model: gpt-oss:20b

• Purpose: Generates creative, humorous tweets based on the user-provided topic.

• Strengths:

  • Strong generative creativity
  • Produces multiple candidate tweets
  • Explores diverse humorous angles

generator_model = ChatOllama(model="gpt-oss:20b")

---

2. Evaluator Agent

• Model: llama3.1:latest

• Purpose: Critically evaluates generated tweets based on predefined quality metrics.

• Evaluation Criteria:

  • Humor
  • Virality potential
  • Clarity
  • Brevity
  • Relevance to topic

evaluator_model = ChatOllama(model="llama3.1:latest")

---

3. Optimizer Agent

• Model: llama3.2:3b

• Purpose: Refines tweets using evaluator feedback, improving punchlines, wording, and structure.

• Why a smaller model?

  • Faster iterations
  • Lower compute cost
  • Ideal for targeted rewriting and fine-tuning

optimizer_model = ChatOllama(model="llama3.2:3b")

---

🔁 LangGraph Workflow

The entire system is orchestrated using LangGraph, enabling a clean, stateful, and iterative agent flow.

Workflow Steps:

1. Input Topic
2. Tweet Generation
3. Tweet Evaluation
4. Tweet Optimization
5. Quality Check
6. Repeat until optimal tweet is achieved

graph TD
    A[User Topic] --> B[Generator Agent]
    B --> C[Evaluator Agent]
    C --> D[Optimizer Agent]
    D --> C
    C -->|Meets Quality Threshold| E[Final Tweet]

Loading

---

🛠️ Tech Stack

Component	Technology
Agent Framework	LangGraph
LLM Provider	Ollama
Generator Model	gpt-oss:20b
Evaluator Model	llama3.1
Optimizer Model	llama3.2:3b
Language	Python

---

⚙️ Installation & Setup

Prerequisites

• Python 3.9+
• Ollama installed locally
• Required models pulled via Ollama

ollama pull gpt-oss:20b
ollama pull llama3.1
ollama pull llama3.2:3b

---

▶️ Usage

python main.py --topic "AI replacing human jobs"

Example Output

“AI won’t take your job. It’ll just ask for your help… then quietly stop replying.”

---

🎯 Benefits

✅ For Developers

• Demonstrates agentic AI design patterns
• Real-world use of LangGraph
• Easy to extend with more agents or evaluation criteria

✅ For Content Creators

• Generates high-quality, funny tweets
• Reduces creative fatigue
• Increases engagement potential

✅ For Researchers

• Experiment with self-improving LLM systems
• Compare multi-model collaboration vs single-model prompting

---

⭐ Acknowledgements

• LangGraph for agent orchestration
• Ollama for local, open-source LLM inference
• The open-source LLM community ❤️

---