A Streamlit application that combines RAGAnything (multimodal RAG) with DeepAgent to automatically generate professional Business Requirements Documents (BRDs) from any uploaded files.
Uploaded Files (PDF, DOCX, PPTX, XLSX, TXT, MD, PNG, JPG, …)
│
▼
┌─────────────────────────────────────┐
│ RAGAnything Pipeline │
│ ┌───────────────────────────────┐ │
│ │ MinerU Document Parser │ │ ← Extracts text, tables,
│ │ (PDF / Office / Image / …) │ │ equations, images
│ └──────────────┬────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────┐ │
│ │ Multimodal Content Analysis │ │ ← Vision LLM for images,
│ │ (Image / Table / Equation) │ │ table interpreter, LaTeX
│ └──────────────┬────────────────┘ │
│ ▼ │
│ ┌───────────────────────────────┐ │
│ │ LightRAG Knowledge Graph │ │ ← Entity extraction +
│ │ (Vector + Graph Index) │ │ cross-modal relationships
│ └──────────────┬────────────────┘ │
└─────────────────┼───────────────────┘
│ hybrid / local / global query
▼
RAG Context (rich text)
│
▼
┌─────────────────────────────────────┐
│ DeepAgent │
│ model: gemini / gpt-4o / claude │
│ system: Senior Business Analyst │
│ input: RAG context + user query │
│ output: Full BRD in Markdown │
└─────────────────────────────────────┘
# Install LibreOffice for Office document support (macOS / Linux)
brew install --cask libreoffice # macOS
sudo apt-get install libreoffice # Ubuntu/Debian
# Install Python packages
pip install -r requirements.txt
# Verify MinerU
mineru --versioncp env.example .env
# Edit .env and fill in your keysOr enter them directly in the sidebar when the app is running.
streamlit run app.pyOpen http://localhost:8501 in your browser.
- Upload files — any combination of PDFs, Word docs, presentations, spreadsheets, images, markdown files, etc.
- Describe your needs — e.g. "Generate a BRD for a mobile banking app based on these requirements docs. Focus on authentication and payments."
- Pick BRD sections in the sidebar (defaults are pre-selected).
- Click Generate BRD — the pipeline will:
- Parse and index all files with RAGAnything
- Query the knowledge base for relevant context
- Feed context + query to DeepAgent
- Render the complete BRD in the right panel
- Download the Markdown BRD.
| Format | Extension(s) | Notes |
|---|---|---|
| Text + images + tables | ||
| Word | .docx, .doc | Requires LibreOffice |
| PowerPoint | .pptx, .ppt | Requires LibreOffice |
| Excel | .xlsx, .xls | Requires LibreOffice |
| Images | .jpg, .png, .bmp, .tiff, .webp | Vision LLM analysis |
| Text/Markdown | .txt, .md | Direct text ingestion |
| Provider | Models |
|---|---|
| gemini-2.0-flash-preview, gemini-1.5-pro | |
| OpenAI | gpt-4o, gpt-4o-mini |
| Anthropic | claude-opus-4-6, claude-sonnet-4-6 |
Note: RAGAnything's internal LLM and embedding calls always use OpenAI (gpt-4o-mini + text-embedding-3-large). An OpenAI API key is therefore required regardless of which DeepAgent model you choose.
deep-brd-agent/
├── app.py # Streamlit UI
├── rag_pipeline.py # RAGAnything + DeepAgent integration
├── requirements.txt # Python dependencies
├── env.example # Environment variable template
└── README.md # This file
| Option | Description |
|---|---|
| DeepAgent Model | LLM model for BRD generation |
| API Keys | Provider keys (Google / OpenAI / Anthropic) |
| Query Mode | RAG retrieval strategy (hybrid recommended) |
| Embedding Model | OpenAI embedding model |
| Process Images | Enable vision analysis for image content |
| Process Tables | Enable structured data interpretation |
| Process Equations | Enable LaTeX equation parsing |
| BRD Sections | Sections to include in the output document |