polojuan/agentic-workflows
0
1---2title: Agentic Workflows3emoji: 🤖4colorFrom: blue5colorTo: gray6sdk: streamlit7sdk_version: 1.49.18app_file: app.py9pinned: false10license: mit11short_description: 'Multi-Agent Systems for Research and Database Analysis'12---13 14# 🤖 Agentic Workflow Platform15 16An intelligent multi-agent platform providing specialized AI assistants for different workflows—from academic research to medical evidence gathering to database analysis. Choose your assistant and let AI agents handle the complexity while you focus on insights.17 18192021 22## 🌟 Features23 24### **🤖 Fully Autonomous**25- Zero manual intervention required26- Intelligent task planning and decomposition27- Self-organizing multi-agent coordination28 29### **🎯 Evidence-Based**30- Searches only credible, peer-reviewed sources31- Medical databases (PubMed, Cochrane Library)32- Academic databases (arXiv, Wikipedia, Tavily)33- APA format citations34 35### **📝 Complete Reports**36- Structured markdown output37- Introduction, findings, and conclusions38- Comprehensive reference sections39- Publication-ready format40 41## 🏗️ Architecture42 43### Research Workflows (5-Agent System)44 45The research assistants use a **5-agent workflow** that works collaboratively:46 471. **Planner Agent** - Breaks down research topics into actionable steps482. **Research/Medical Agent** - Searches credible databases and sources493. **Writer Agent** - Drafts well-structured summaries and reports504. **Editor Agent** - Reviews, critiques, and refines content515. **Execution Agent** - Orchestrates the entire workflow52 53### Database Workflow (Text-to-SQL)54 55The OpenRCA assistant uses an **iterative refinement workflow**:56 571. **Database Agent** - Converts natural language questions to SQL queries582. **Query Executor** - Runs SQL against the database593. **Result Evaluator** - Reviews query results for accuracy604. **Query Refiner** - Iteratively improves SQL based on feedback615. **Result Interpreter** - Translates data into natural language answers62 63## 📱 Applications64 65The application includes four pages accessible via the Streamlit sidebar:66 67### 🏠 **Landing Page**68- Overview of the agentic workflow platform69- Choose between Research Assistant, Clinical Evidence, or OpenRCA70- Key features and capabilities71- Navigation to specialized assistants72 73### ✍️ **Research Assistant** (General Academic & Scientific)74 75**Ideal for:**76- Academic research papers77- Technical documentation78- Scientific literature reviews79- General knowledge gathering80 81**Data Sources:**82- 📚 **arXiv** - Academic papers and preprints83- 🌐 **Tavily** - General web search84- 📖 **Wikipedia** - Encyclopedic knowledge85 86**Typical Use Cases:**87- Computer Science & AI research88- Physics & Mathematics89- Engineering topics90- General academic inquiries91 92**Workflow Time:** 5-10 minutes93 94### 🔍 **Dr. ResearchRx** (Evidence-Based Medical Research)95 96**Ideal for:**97- Clinical research98- Medical literature reviews99- Evidence-based medicine100- Healthcare decision support101 102**Data Sources:**103- 🩺 **PubMed** - Medical research papers & clinical studies104- 📊 **Cochrane Library** - Systematic reviews & meta-analyses105 106**Typical Use Cases:**107- Disease research & treatment options108- Drug efficacy & safety studies109- Clinical guidelines & protocols110- Medical condition analysis111 112**Workflow Time:** 2-3 minutes113 114### 🔧 **OpenRCA** (Root Cause Analysis Database Assistant)115 116**Ideal for:**117- Equipment failure analysis118- Root cause investigation119- Corrective action tracking120- Maintenance data exploration121 122**Features:**123- 🗣️ **Natural Language Queries** - Ask questions in plain English124- 🔄 **Smart SQL Generation** - AI converts questions to accurate SQL125- ✅ **Iterative Refinement** - Automatically improves query accuracy126- 📊 **Database Browser** - Interactive table exploration with filtering127- 📥 **Export Data** - Download filtered results as CSV128 129**Typical Use Cases:**130- Finding most expensive failures by equipment/area131- Identifying common root causes and failure patterns132- Analyzing downtime and maintenance costs133- Reviewing corrective actions and their effectiveness134 135**Database Schema:**136- Equipment failures with timestamps and costs137- Root cause analysis records138- Corrective actions and prevention measures139- Asset hierarchy (Asset → Area → Equipment)140 141**Workflow Time:** 1-2 minutes142 143## 🚀 Getting Started144 145### Prerequisites146 147- Python 3.8 or higher148- OpenAI API key149- Tavily API key150 151### Installation152 1531. **Clone the repository:**154```bash155git clone https://github.com/palscruz23/agentic-workflow.git156cd agentic-workflow157```158 1592. **Install dependencies:**160```bash161pip install -r requirements.txt162```163 1643. **Set up environment variables:**165 166Create a `.env` file in the root directory:167```bash168OPENAI_API_KEY=your_openai_api_key_here169TAVILY_API_KEY=your_tavily_api_key_here170```171 1724. **Run the application:**173```bash174streamlit run app.py175```176 177The application will open in your default browser at `http://localhost:8501`178 179### 🐳 Docker Deployment180 181You can also run the application using Docker for a containerized deployment:182 183**Option 1: Using Docker Compose (Recommended)**184 1851. **Create `.env` file** with your API keys:186```bash187OPENAI_API_KEY=your_openai_api_key_here188TAVILY_API_KEY=your_tavily_api_key_here189```190 1912. **Build and run with Docker Compose:**192```bash193docker-compose up -d194```195 1963. **Access the application:**197Open your browser at `http://localhost:8501`198 1994. **Stop the container:**200```bash201docker-compose down202```203 204**Option 2: Using Docker directly**205 2061. **Build the Docker image:**207```bash208docker build -t agentic-workflow .209```210 2112. **Run the container:**212```bash213docker run -d \214 -p 8501:8501 \215 -e OPENAI_API_KEY=your_openai_api_key_here \216 -e TAVILY_API_KEY=your_tavily_api_key_here \217 --name agentic-research \218 agentic-workflow219```220 2213. **View logs:**222```bash223docker logs -f agentic-research224```225 2264. **Stop the container:**227```bash228docker stop agentic-research229docker rm agentic-research230```231 232## 📂 Project Structure233 234```235agentic-workflow/236├── app.py # Landing page (main entry point)237├── About.py # About page with workflow details238├── pages/239│ ├── 1_Research Assistant.py # General academic research240│ ├── 2_Clinical Evidence.py # Medical research241│ └── 3_OpenRCA.py # Database assistant for RCA242├── agents/243│ ├── planner_agent.py # Task planning and decomposition244│ ├── research_agent.py # General research (arXiv, Tavily, Wikipedia)245│ ├── medical_agent.py # Medical research (PubMed, Cochrane)246│ ├── writer_agent.py # Content generation247│ ├── editor_agent.py # Content review and refinement248│ ├── execution_agent.py # Workflow orchestration249│ └── database_agent.py # Text-to-SQL with iterative refinement250├── tools/251│ ├── research_tools.py # arXiv, Tavily, Wikipedia search tools252│ └── medical_tools.py # PubMed, Cochrane search tools253├── utils/254│ └── database.py # Database utilities and schema255├── data/256│ └── rca_data.db # SQLite database for RCA257├── requirements.txt # Python dependencies258├── Dockerfile # Docker image configuration259├── docker-compose.yml # Docker Compose configuration260├── .dockerignore # Docker build ignore file261├── .env # Environment variables (create this)262└── README.md # This file263```264 265## 🔧 Configuration266 267### Environment Variables268 269- `OPENAI_API_KEY` - Your OpenAI API key (required)270- `TAVILY_API_KEY` - Your Tavily API key (required)271- `DLAI_TAVILY_BASE_URL` - Optional custom Tavily base URL272 273### Model Configuration274 275By default, the application uses `gpt-4o-mini`. You can modify this in `app.py`:276 277```python278if "model" not in st.session_state:279 st.session_state.model = "gpt-4o-mini" # Change this280```281 282## 💡 Usage Examples283 284### Example 1: General Academic Research285 286**Topic:** "Transformer architecture in neural networks"287 288**Agent Workflow:**2891. **Planner:** Breaks down into 5 steps (search arXiv, search Wikipedia, draft report, edit, finalize)2902. **Research Agent:** Searches arXiv for papers on transformers, queries Wikipedia for background2913. **Writer Agent:** Drafts comprehensive report with findings2924. **Editor Agent:** Reviews and improves clarity, structure2935. **Execution Agent:** Delivers final markdown report with references294 295**Output:** Complete research report with introduction, key findings, and APA citations296 297### Example 2: Medical Research298 299**Topic:** "Latest treatments for Type 2 Diabetes"300 301**Agent Workflow:**3021. **Planner:** Creates medical research plan3032. **Medical Agent:** Searches PubMed for clinical studies, Cochrane for systematic reviews3043. **Writer Agent:** Compiles evidence-based summary3054. **Editor Agent:** Refines medical terminology and accuracy3065. **Execution Agent:** Produces clinical evidence report307 308**Output:** Evidence-based medical report with peer-reviewed sources309 310### Example 3: Database Analysis (OpenRCA)311 312**Question:** "What are the top 3 most expensive failures?"313 314**Agent Workflow:**3151. **Database Agent:** Converts question to SQL query3162. **Query Executor:** Runs initial SQL query against RCA database3173. **Result Evaluator:** Reviews if results correctly answer the question3184. **Query Refiner:** Improves SQL to include relevant columns and fix aggregations3195. **Result Interpreter:** Translates data into natural language answer320 321**Output:** "The top 3 most expensive failures are: 1) Mine B Concentrate Storage area with an impact of 3,492,427 AUD due to improper tensioning; 2) Mine A Haul Road area with an impact of 3,344,968 AUD due to software malfunction; and 3) Mine A Water Treatment area with an impact of 3,135,004 AUD due to thermal stress."322 323## 🛠️ Development324 325### Adding New Tools326 327To add a new research tool:328 3291. Create the tool function in `tools/research_tools.py` or `tools/medical_tools.py`:330```python331def new_search_tool(query: str, max_results: int = 5) -> list[dict]:332 # Implementation333 return results334```335 3362. Define the tool specification:337```python338new_tool_def = {339 "type": "function",340 "function": {341 "name": "new_search_tool",342 "description": "Description of what this tool does",343 "parameters": {...}344 }345}346```347 3483. Add to the appropriate agent's tool mapping349 350### Customizing Agents351 352Each agent can be customized by modifying the system prompts in the respective agent files:353- `agents/planner_agent.py` - Planning strategy354- `agents/research_agent.py` - Research approach355- `agents/writer_agent.py` - Writing style356- `agents/editor_agent.py` - Review criteria357 358## 🧪 Testing359 360Run the application locally to test:361```bash362streamlit run app.py363```364 365Navigate to each page to ensure:366- ✅ Landing page displays correctly with all three assistants367- ✅ Research Assistant can search arXiv, Tavily, Wikipedia368- ✅ Clinical Evidence can search PubMed, Cochrane369- ✅ OpenRCA can query database and refine SQL queries370- ✅ OpenRCA database browser can filter and export data371- ✅ Reports are generated with proper formatting372 373## 🐛 Troubleshooting374 375### Common Issues376 377**Issue: Import Error for agents**378```379ImportError: cannot import name 'planner_agent' from 'agents.planner_agent'380```381**Solution:** Ensure all agent files have proper OpenAI client initialization inside functions, not at module level.382 383**Issue: API Key Errors**384```385ValueError: TAVILY_API_KEY not found in environment variables386```387**Solution:** Create `.env` file with required API keys.388 389**Issue: Session State Errors**390```391AttributeError: st.session_state has no attribute 'client'392```393**Solution:** The app initializes session state in `init_chatbot()`. Ensure it's called before using agents.394 395## 🤝 Contributing396 397Contributions are welcome! Here's how:398 3991. Fork the repository4002. Create a feature branch (`git checkout -b feature/amazing-feature`)4013. Commit your changes (`git commit -m 'Add amazing feature'`)4024. Push to the branch (`git push origin feature/amazing-feature`)4035. Open a Pull Request404 405## 📄 License406 407This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.408 409## 🙏 Acknowledgments410 411- Built with [Streamlit](https://streamlit.io/)412- Powered by [OpenAI GPT-4o-mini](https://openai.com/)413- Research tools: [arXiv](https://arxiv.org/), [PubMed](https://pubmed.ncbi.nlm.nih.gov/), [Cochrane Library](https://www.cochranelibrary.com/)414- Web search: [Tavily](https://tavily.com/)415 416## 📞 Support417 418- **Issues:** [GitHub Issues](https://github.com/palscruz23/agentic-workflow/issues)419- **Repository:** [GitHub Repository](https://github.com/palscruz23/rul-estimation)420 421---422 423**Built with Claude, OpenAI, Streamlit | Powered by Multi-Agent AI** 🤖424