ClinicFlow Agent

Autonomous AI agent for clinics that reads patient enquiries, drafts personalized replies, gets human approval, sends automatically, and schedules 24-hour follow-ups. Zero manual effort.

Comment

Inspiration

Every day, clinics lose patients not because they don't care — but because they are too busy to reply quickly. A patient sends a message at 7pm. Nobody replies until the next morning. By then, that patient has booked somewhere else. I built ClinicFlow Agent to close that gap permanently.

What it does

ClinicFlow Agent is a fully autonomous patient communication system for appointment-based clinics. When a patient sends an enquiry, the agent:

  1. Reads and classifies the message — understanding intent, urgency, and topics
  2. Drafts a warm, personalised reply using Qwen Cloud AI
  3. Presents the reply for one-click human approval
  4. Sends the approved reply automatically
  5. Schedules a 24-hour follow-up if the patient does not respond Zero manual effort. Zero missed leads.

How I built it

  • Python backend with Flask web framework
  • Qwen Cloud API (qwen-plus model) for message classification and reply generation
  • Clean web dashboard with real-time agent workflow display
  • Human-in-the-loop approval checkpoint built into the workflow
  • Deployed on Alibaba Cloud ECS

Challenges I ran into

Getting the Qwen Cloud API to return clean JSON classification results consistently required careful prompt engineering. I also had to design the human approval checkpoint so it feels natural in a web interface rather than a terminal prompt.

Accomplishments that I am proud of

Building a fully working autonomous AI agent from scratch in under 72 hours - from a Python terminal script to a professional web dashboard running on Qwen Cloud. The human-in-the-loop checkpoint was the part I am most proud of. It would have been easy to make the agent send replies without any review. But real clinics need to trust the system before they hand it full control. That checkpoint - one click to approve, edit, or reject - makes ClinicFlow Agent something a real business owner would actually use. I am also proud of building this as a solo developer while managing multiple other commitments. Every line of code, every design decision, every debugging session - done alone, in public, documented day by day.

What I learned

I learned how to build a production-ready autonomous agent that handles real-world, messy inputs - not just clean test data. I also learned the importance of keeping the human in control at critical decision points.

What's next for ClinicFlow Agent

  • WhatsApp and email integration for real message channels
  • Multi-clinic dashboard for managing several locations
  • Memory layer so the agent learns each clinic's tone and preferences over time

Built With

Share this project:

Updates