CAFBrain

Capital Area Food Bank's Generative AI Assistant for Fundraising and Communications Teams

CAFBrain is an AI-powered Streamlit application designed to transform, query, and refine fundraising and communication materials. Built with LLMs, RAG (Retrieval-Augmented Generation), and multimodal input handling, it empowers users to generate grant proposals, blogs, presentations, and reports using their own source files—with full human-in-the-loop refinement and download support.

---

🚀 Features

• LLM-Powered Generation: Uses large language models (LLMs) to generate high-quality collateral.
• RAG-Based Querying: Ask questions and extract relevant insights from uploaded source documents.
• Multimodal Input Support: Accepts PDFs, PPTX, DOCX, TXT, and Images (with OCR for text extraction).
• Human-in-the-Loop Refinement: Users can iteratively refine outputs, upload additional supporting files, and regenerate content.
• Template-Based Injection: Outputs are fitted into user-provided templates (PPTX, PDF, DOCX).
• Streamlit Web App Interface: Fully interactive frontend to upload files, select modes, refine results, and download outputs.
• Multiple Output Formats: Download generated content as PowerPoint presentations, PDF reports, or HTML blogs.
• Session Persistence: Maintains session state for smooth transitions across multiple generations and refinements.

---

🛠️ Tech Stack

• Python 3.11+
• Streamlit
• LangChain
• Langgraph
• OpenAI API
• Unstructured (document parsing)
• PyMuPDF, pytesseract (image OCR)
• python-pptx (PowerPoint generation)
• FPDF (PDF generation)
• FastAPI (optional extension for backend APIs)

---

⚙️ How to Run Locally

1. Clone the Repository

git clone https://github.com/CAFBrain-Project/CAFBrain.git
cd CAFBrain

2. Install Dependencies

pip install -r requirements.txt

3. Set OpenAI API Key

export OPENAI_API_KEY=your-key-here

4. Launch the Streamlit App

streamlit run app.py

---

🧠 Workflow Overview

1. Upload Source Files

   • Supported types: PDF, PPTX, DOCX, TXT, Images.

2. Choose Action Mode

   • Query: Ask questions to retrieve specific information from your uploads.
   • Generate New Content: Create new grant proposals, blog posts, presentations, and reports.
   • Refinement: Refine existing generated content with extra supporting files.

3. Optionally Upload Extra Files

   • During refinement, users can add more documents to enhance outputs.

4. Human-in-the-Loop Interaction

   • Edit, regenerate, or query again for iterative improvement.

5. Download Outputs

   • Supported formats: PPTX (Presentations), PDF (Reports/Proposals), HTML (Blogs).

---

✨ Example Use Cases

Scenario	Action	Output
Past grants + Strategic Plan	Generate New Grant Proposal	Grant proposal emphasizing Food-as-Medicine leadership
Grant report + Interview Transcript	Generate Blog Post	Client-focused blog with embedded quotes
Strategic Plan + Budget Documents	Generate Executive Presentation	Board-ready strategic deck
New client interviews + prior blogs	Refine Existing Content	Updated blog with fresh testimonials

---

📌 Future Enhancements

• Drag-and-drop template selection
• Semantic search for deeper RAG-based querying
• In-app live editing of generated content
• Audit trail and versioning for refinements

---

🤝 Acknowledgements

This project is developed for the Capital Area Food Bank by us.
Special thanks to the CAFB Fundraising, University of Maryland and Communications teams for continuous feedback, insights, and support.

---

📜 License

This project is currently licensed for internal use by Capital Area Food Bank.