Agents for Construction Scheduling Management
I recently tested an AI agent that semi-automates scheduling for contractors. This agent:
- Handles messages from supervisors, workers, and customers
- Schedules jobs
- Assigns workers
- Notifies supervisors and workers of schedule changes
- Streamlines the scheduling process and reduces manual effort
Technologies Used
- OpenAI & Anthropic API: Natural language understanding and generation
- LlamaIndex: Retrieval-augmented generation systems around private data
- Langchain: Extending LlamaIndex with context-aware tools
- FastAPI: APIs to serve custom company data to the models
- Streamlit: Interactive web application to showcase and manage scheduling tasks
- Playwright: Testing interactions to ensure the reliability of the AI agent
Implementation Steps
Infrastructure Setup with FastAPI
Developed backend infrastructure using FastAPI to connect AI models with the frontend application, handling API requests and integrating with other system components.
Setup RAG and Context-Aware Tools with LlamaIndex and Langchain
Built a RAG system with LlamaIndex for AI access to private project data. Used Langchain to add context-aware functionalities, improving AI's understanding and response to scheduling requests.
Setup OpenAI & Anthropic API for NLP
Connected both OpenAI and Anthropic APIs for NLP/text generation, ensuring wide-ranging customer inquiry handling and precise scheduling responses, with added redundancy for API disruptions.
Developing the Interface with Streamlit
Created a web app using Streamlit, allowing supervisors to input project details, view schedules, and manage assignments.
Testing with Playwright
Implemented automated testing using Playwright to simulate real-world interactions, verifying AI's accuracy in processing scheduling requests and assigning tasks (achieving ~90% success rate).
Setting Up Document Retrieval with LlamaIndex
Utilized LlamaIndex to manage data from project documents and communications, ensuring agent access to worker availability, job specifications, and timelines.
Upcoming Developments
- Voice access to the agent and SMS/calling integration
- LangChain middle-agent for filtering non-conforming or unauthorized submissions
Future Feature Ideas
- Voice Command Capabilities: Integrate voice recognition for supervisors to schedule and manage jobs using voice commands.
- Real-Time Updates and Notifications: Implement push notifications or SMS alerts for instant communication of schedule changes to supervisors and workers.