mussaratshamsher/digital-fte

Digital FTE Factory is a platform for building autonomous AI agents meant to take over manual tasks like CRM data entry and other disconnected office workflows. The idea is that instead of a human repeating the same data entry and coordination tasks across separate tools, specialized AI agents for sales, project management, and support handle those tasks on their own, connected to a company's CRM and knowledge systems. The frontend is built with Next.js, React, TypeScript, and TailwindCSS, while the backend runs on FastAPI in Python with SQLAlchemy and either PostgreSQL or SQLite for the database, plus Supabase for authentication. User login is handled through JWT tokens with OAuth2. For the AI itself, the project uses Groq's cloud hosting to run the Llama 3.3 70B model, chosen by the authors for its fast response times and lower cost compared to models like GPT-4o or Claude. The system can be deployed with Docker and Kubernetes. When a user triggers a task, an orchestrator built on FastAPI breaks it into smaller sub-tasks, which are then handled by the specialized agents using their own tools, with results logged and saved to the database in real time. The README includes detailed cost estimates, claiming a typical task costs under a cent in token usage and that running the system at volume would cost around $41 a month in AI compute, alongside a comparison table contrasting AI agent costs against a mid-level human employee's salary. The documentation is written from a business and ROI perspective rather than a pure technical reference, with tables breaking down token usage per step and cost per task. It also links to separate documents for system architecture, a demo script, and a deployment guide, all located in a phase-05 folder, suggesting the project has been built and documented in stages.