A modular, agent-based system that crafts polished LinkedIn posts through orchestrated phases: understanding intent, generating a behind-the-scenes story, optimizing hashtags, drafting the final post, and optionally creating a matching image.
I built this system to help people avoid generic AI-sounding posts and produce content that better reflects their voice, intent, and goals. A manager agent coordinates specialist sub-agents so every post feels intentional, professional, and personalized.
Many AI-assisted LinkedIn posts look similar and lose personal voice. Creators need support that improves quality without flattening authenticity.
A Google ADK-based multi-agent workflow where each specialist handles one phase of writing, while a manager agent orchestrates the full pipeline and returns a cohesive final post.
Phase 1: Gather user intent and context
Phase 2: Generate an engaging behind story
Phase 3: Suggest relevant, discoverable hashtags
Phase 4: Craft a complete polished LinkedIn post
Phase 5 (optional): Generate an image prompt and asset
The manager agent delegates tasks to focused sub-agents for story generation, hashtag selection, final post composition, and optional image creation.
This modular setup keeps each responsibility clear, improves maintainability, and makes it easy to iterate on one stage without rewriting the entire system.
Runtime: Python-based service orchestrated with Google ADK, using Gemini APIs for content generation and optional image workflows with Cloudinary integration.
# Clone the repository git clone https://github.com/sujeetgund/linkedin-post-generator-agent.git cd linkedin-post-generator-agent # Python setup python -m venv .venv .venv\Scripts\activate pip install -r requirements.txt # Configure secrets cd linkedin_post_agent copy .env.example .env # Add required keys in .env # GOOGLE_API_KEY=... # CLOUDINARY_CLOUD_NAME=... # CLOUDINARY_API_KEY=... # CLOUDINARY_API_SECRET=... # Run the agent system python -m linkedin_post_agent # Optional: ADK web UI adk web
Manager agent orchestrates specialist sub-agents and combines outputs into a final post, with optional image generation.
The phased workflow creates clearer, more compelling posts by separating idea capture, storytelling, and final composition.
Individual sub-agents can be improved independently, making the system easier to maintain and extend over time.
I enjoy building practical multi-agent systems for real-world creator workflows. Let's connect and discuss collaboration.