CampaignMaker AI

Create Social Media Campaigns with single prompt. Fastapi, Next.js, ChromaDB, MongoDB, Google Agent Development Kit.

Multi-Agent Orchestration

The core strength of CampaignMaker lies in its multi-agent system, where specialized AI agents collaborate to deliver a cohesive marketing campaign.

• Root Agent (root_agent): The primary orchestrator. It interacts with the user, understands the overall goal, initializes the necessary context states, and delegates tasks to the appropriate sub-agents.
• Campaign Maker (campaign_maker): A LoopAgent responsible for strategy. It loops between generating and reviewing to ensure high quality:
  • Campaign Outline Generator (campaign_outline_generator): Drafts the initial campaign strategy and post outlines.
  • Campaign Outline Reviewer (campaign_outline_reviewer): Critiques the drafted strategy and refines it. This loop continues until a satisfactory strategy is achieved (up to 3 iterations).
• Content Generator Agent (content_generator_agent): A router agent that delegates content creation based on the target social media platform.
  • Platform Content Generators (e.g., twitter_content_generator, linkedin_content_generator, instagram_content_generator): Orchestrates text and media generation for specific platforms.
  • Text Generator (twitter_post_text_generator): Crafts the actual copy for the posts.
  • Graphic Generator (svg_graphic_generator): Produces SVG code to create visual illustrations for the posts.

State Management & Data Persistence

To ensure seamless collaboration between agents and prevent context loss, CampaignMaker employs a robust state management system.

• Context Narrowing: By breaking down the task, specific agent receives only the specific context (state) it needs rather than the entire conversation history.
• Persistence: Utilizes callbacks (like initialize_state) and tools (like save_state) tied to the Root Agent. This allows to track various states—such as campaign_strategy, approved_outlines, and generated_content—ensuring data is persisted across agent handoffs and displayed correctly in the frontend UI.

Document context (ChromaDB RAG)

CampaignMaker can ground a campaign in the user's brand material. Use the paperclip in the chat to upload up to five PDF, DOCX, TXT, Markdown, or CSV files (15 MB each by default). The API extracts text, splits it into overlapping chunks, creates Gemini embeddings, and persists the chunks in ChromaDB. The root agent then has a scoped search_uploaded_documents tool it uses to retrieve relevant brand facts while planning and writing campaign content.

The knowledge base is stored at backend/data/chroma by default (configure with CHROMA_PERSIST_DIRECTORY) and every chunk is filtered by the authenticated user ID before retrieval. Original uploads are not stored after their text is indexed. Scanned PDFs need OCR before upload.

Required backend configuration:

GOOGLE_API_KEY=your-gemini-api-key
# Optional
CHROMA_PERSIST_DIRECTORY=./data/chroma
RAG_EMBEDDING_MODEL=gemini-embedding-001
RAG_CHUNK_SIZE=1200
RAG_CHUNK_OVERLAP=200

Known Deficiencies & Shortcomings

While CampaignMaker is highly capable, there are a few current limitations:

• SVG Image Realism: The svg_graphic_generator produces vector illustrations. While useful for simple graphics and layouts, SVGs generated purely by LLMs lack the photorealism and complex artistry seen in image-based models (nano-banana). They are functional, but not "worthwhile".
• Schema Validation Nuances: Because we rely on strict Pydantic JSON schemas to pass data between agents, occasional validation errors can occur if an agent hallucinates outside the expected JSON format.

Future Additions

• Automated Social Media Posting:To implement direct-to-platform posting using OAuth 2.0 authorization for differnt social media platform. This will allow users to securely grant permissions for CampaignMaker to publish directly to Twitter, LinkedIn, and Instagram without leaving the app.
• Advanced Media Integrations: Replacing or augmenting the SVG generator with an external image generation API for high-fidelity raster images.
• Enhanced Analytics UI: Visualizing the predicted engagement of campaigns based on the strategy reviewed by the agents.