Automate Subcontractor Communication and Task Tracking with Custom AI

Construction firms often rely on a combination of a project management system like Procore, email, and Excel spreadsheets. Procore is a powerful system of record, but its communication modules can be rigid. As a result, subcontractors default to what is easiest for them: sending an RFI as a PDF attached to an email. This immediately breaks any structured workflow the general contractor tried to establish.

A project manager at a 20-person firm receives an RFI via email from an electrical sub. To process it, the PM must open the email, download the PDF, identify the project, log into Procore, manually create a new RFI record, copy-paste the details, upload the PDF, and finally assign it to the right architect. This manual data entry takes 5-10 minutes per RFI. Across five active projects with dozens of RFIs each week, the PM spends hours on administrative work instead of managing the project.

The structural problem is that project management tools are designed as databases, not as intelligent intake systems. They expect structured data entered through their own forms. They have no native capability to read an unstructured email or a poorly formatted PDF from a subcontractor and extract the necessary information. You can ask subs to use a portal, but they never do consistently. This gap between real-world communication and system requirements forces your most expensive employees into manual data entry roles.

The first step would be an audit of your current communication flows. We would map how RFIs, submittals, and change orders arrive from your top 10-15 subcontractors. We analyze the different formats (email body, PDF, Word doc) and identify the key pieces of information that need to be extracted for each document type. You receive a data map outlining the extraction logic before any code is written.

The technical approach uses an AWS Lambda function that triggers on incoming emails to a dedicated address. The Claude API parses the email body and any attachments to extract project name, sender, and the specific request. We use the Claude API because its large context window can handle multi-page PDFs and complex contract language, a problem where we have built similar document processing pipelines for financial services. A FastAPI service then validates this extracted data and writes it to a Supabase database, creating a clean, structured record of every communication.

The delivered system integrates directly with your primary project management tool or a custom Supabase dashboard. Your project managers would receive an alert with a link to the categorized RFI, ready for their review and assignment, not data entry. You receive the full Python source code, an architecture diagram, and a runbook for maintenance. The system would process a typical RFI email in under 60 seconds.