CareForMe - Agents for Humans
Inovating Clinic Operations with Strands Agents and Amazon Bedrock | Agents for Humans
1. Introduction: The Real Headache in Local Clinics
There is a timeless saying that everyone knows: "Health is wealth." We all want prompt access to care when we feel unwell, and we rely on
our local community clinics to keep our families healthy.
Yet behind the scenes, these clinics face a quiet crisis happening every single day right at the front desk: administrative exhaustion.
When people talk about AI in healthcare, you usually hear about futuristic ideas: like AI discovering new drugs or assisting surgeons in the
operating room. Those are great, but they overlook the everyday friction regular people face. If you have ever visited a small clinic, you
know the front-desk staff are almost always overwhelmed. They are answering ringing phones, checking in walk-ins, scheduling appointments, chasing down no-shows, and manually typing out reminders.
know the front-desk staff are almost always overwhelmed. They are answering ringing phones, checking in walk-ins, scheduling appointments, chasing down no-shows, and manually typing out reminders.
This creates real problems:
• Missed Appointments: Between 15% and 30% of patients simply miss their visits. That hurts the clinic financially, but worse, it means a patient misses out on preventive care.
• Tired Staff: Instead of greeting patients with a warm smile, staff spend hours playing frustrating "phone tag."
• Too Many Screens: Most software just gives staff another complicated dashboard with 50 buttons to click.
When AWS launched the Agents for Humans Hackathon, the challenge resonated with us: Build an AI agent that actually does real work for real people not just a chatbot that talks.
• Missed Appointments: Between 15% and 30% of patients simply miss their visits. That hurts the clinic financially, but worse, it means a patient misses out on preventive care.
• Tired Staff: Instead of greeting patients with a warm smile, staff spend hours playing frustrating "phone tag."
• Too Many Screens: Most software just gives staff another complicated dashboard with 50 buttons to click.
When AWS launched the Agents for Humans Hackathon, the challenge resonated with us: Build an AI agent that actually does real work for real people not just a chatbot that talks.
That is why we built CareForMe.
CareForMe is an autonomous, safety-bounded assistant for clinics built with the Strands Agents SDK, Amazon Bedrock (Nova Lite), Amazon
DynamoDB, and Amazon Cognito. Instead of just giving advice, CareForMe inspects the clinic schedule, texts patients over SMS or WhatsApp,
books appointments, coordinates reschedules, and updates clinic records automatically.
CareForMe is an autonomous, safety-bounded assistant for clinics built with the Strands Agents SDK, Amazon Bedrock (Nova Lite), Amazon
DynamoDB, and Amazon Cognito. Instead of just giving advice, CareForMe inspects the clinic schedule, texts patients over SMS or WhatsApp,
books appointments, coordinates reschedules, and updates clinic records automatically.
And most importantly: it has strict safety guardrails so it never acts like a doctor or gives medical advice.
Here is the story of how we built it.
──────
──────
2. A Crucial Rule: Help with Admin, Never Play Doctor
The biggest risk with AI in healthcare is letting it do things it shouldn't. If an AI tries to tell someone what medicine to take or
diagnose an illness, it can be dangerous.
diagnose an illness, it can be dangerous.
Right from day one, we made a simple rule:
│ CareForMe is an administrative assistant, NOT a doctor.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
┌────────────────────────┐
│ Patient Texts │
│ the Clinic Number │
└───────────┬────────────┘
│
┌──────────────────────┴──────────────────────┐
▼ ▼
[ Simple Booking Request ] [ Health or Symptom Issue ]
- "1 to confirm" - "I have bad chest pain"
- "2 to reschedule" - "What pills should I take?"
- "3 to cancel" - "I have a high fever"
│ │
▼ ▼
[ AI Handles It Automatically ] [ STOP! Safety Rule Kicks In ]
- Updates the database - Refuses to give medical advice
- Frees up the slot - Tells patient to get urgent help
- Confirms with patient - Creates an alert on dashboard
- Saves in the audit log for human staff to take overHere is how we made sure the AI plays by the rules:
- No Medical Advice: The AI is strictly told it cannot diagnose, prescribe, or interpret tests.
- Instant Alert for Humans: If a patient texts something like "I have terrible chest pain and a high fever," the AI stops right away. It tells the patient to seek medical care and puts a big "Needs Human Review" alert on the clinic's dashboard.
- Audit Trail: Every single thing the AI does reading a file, sending a reminder, or moving an appointment is saved in a log so clinic staff
can inspect it anytime.
──────
3. How It All Connects on AWS
We wanted CareForMe to be fast, reliable, and secure, so we used AWS cloud services to power it.
The Building Blocks:
• The Brain (AI): The Strands Agents SDK connected to Amazon Bedrock using the Amazon Nova Lite model.
• The Database: Amazon DynamoDB to safely store patients, doctors, appointments, and audit logs.
• User Logins: Amazon Cognito so clinic staff can securely log in with passwords and email verification codes.
• The Backend API: Python and FastAPI to handle webhooks and manage clinic logic.
• The Web Dashboard: Next.js, React, and Tailwind CSS so the clinic interface looks modern and easy to use.
• Messaging: Twilio to send real SMS and WhatsApp messages to patients.
──────
• The Database: Amazon DynamoDB to safely store patients, doctors, appointments, and audit logs.
• User Logins: Amazon Cognito so clinic staff can securely log in with passwords and email verification codes.
• The Backend API: Python and FastAPI to handle webhooks and manage clinic logic.
• The Web Dashboard: Next.js, React, and Tailwind CSS so the clinic interface looks modern and easy to use.
• Messaging: Twilio to send real SMS and WhatsApp messages to patients.
──────
4. Building with the Strands Agents SDK & Amazon Nova Lite
To make the AI actually take action, we used the Strands Agents SDK. Strands makes it super simple to turn regular Python functions into
"tools" the AI can use.
Why Amazon Nova Lite on Bedrock?
When an AI is talking to patients or helping staff on a dashboard, three things matter:
- Speed: Nobody wants to wait 10 seconds for a text reply. Nova Lite responds almost instantly.
- Accuracy: When the AI decides to call a function (like booking a slot), it has to provide the right information without making up fake
names or dates. - Cost: Because our agent runs in the background all day, we needed a model that doesn't break the bank.
Amazon Nova Lite checked all three boxes. It is quick, affordable, and followed our tool instructions reliably.
Giving the AI Hands (Tools)
A normal chatbot can only talk. We gave CareForMe "hands" so it could take actions in DynamoDB:
• book_appointment: Checks if the doctor is free, books the slot, and texts the patient a confirmation.
• reschedule_appointment: Moves an appointment to a new date and updates the calendar.
• send_patient_message: Sends an SMS or WhatsApp update to the patient.
• check_past_appointments: Finds people who missed their visits earlier today.
• mark_appointment_status: Updates an appointment to CONFIRMED, CANCELLED, or NO_SHOW.
• escalate_task: Flags a message for human staff when a patient needs medical help.
──────
• reschedule_appointment: Moves an appointment to a new date and updates the calendar.
• send_patient_message: Sends an SMS or WhatsApp update to the patient.
• check_past_appointments: Finds people who missed their visits earlier today.
• mark_appointment_status: Updates an appointment to CONFIRMED, CANCELLED, or NO_SHOW.
• escalate_task: Flags a message for human staff when a patient needs medical help.
──────
5. Three Ways the Agent Works on Its Own
CareForMe doesn't just wait around for someone to click a button. It works in three practical ways:
1. The Automatic Daily Check (Background Routine)
Twice a day, a background script pings the agent with a simple instruction:
│ "Hey CareForMe, do your routine check: look for any past appointments that were missed and follow up."
The AI takes it from there:
- It looks at DynamoDB to find appointments from earlier that never got marked as attended.
- It texts the patient: "We missed you today! Reply 2 if you'd like to reschedule, or call our office."
- It marks the record as NO_SHOW in the database.
Clinic staff didn't have to dial a single phone number.
2. Simple Patient Text Replies
Patients don't need to download another app or remember a password. They just reply to their regular text reminders:
• Reply 1: Confirms their visit (dashboard turns green to show Confirmed).
• Reply 2: Asks to reschedule (alerts staff so a new time can be picked).
• Reply 3: Cancels the visit (instantly opens up that time slot for someone else).
• Reply 2: Asks to reschedule (alerts staff so a new time can be picked).
• Reply 3: Cancels the visit (instantly opens up that time slot for someone else).
If the patient types a message instead of a number, the AI reads it, checks for safety concerns, and responds appropriately.
3. Staff Can Chat with the Agent Directly
Front-desk staff have their own chat window right on the dashboard. Instead of clicking through three different screens, they can just
type:
type:
• "What appointments are coming up tomorrow morning?"
• "Book John Doe with Dr. Sarah Smith for tomorrow at 11:00 AM."
• "Send a text to John reminding him to bring his ID card."
• "Book John Doe with Dr. Sarah Smith for tomorrow at 11:00 AM."
• "Send a text to John reminding him to bring his ID card."
The AI reads the request, runs the proper tool, updates the database, and reports back in plain English.
──────
6. Keeping Clinic Data Safe with DynamoDB
In healthcare, keeping patient info private is a huge deal. You cannot let Clinic A see Clinic B's patients.
We built our DynamoDB tables so that every piece of data is tagged with a unique clinic_id. When staff log in through Amazon Cognito, our
system checks their identity token and ensures the AI can only touch data belonging to their specific clinic.
system checks their identity token and ensures the AI can only touch data belonging to their specific clinic.
We also added a complete Audit Log. Whenever the AI reads a patient, changes an appointment, or sends a text, it writes a timestamped
record. Staff can open the Audit Log page anytime to see exactly what the AI did, when it did it, and why.
record. Staff can open the Audit Log page anytime to see exactly what the AI did, when it did it, and why.
7. Lessons We Learned While Building
Building an AI that actually modifies real database records taught us a lot:
- Dates and Times are Tricky: We noticed AI models can get confused when calculating relative dates (like figuring out what "next Tuesday"
means across timezones). We fixed this by formatting dates clearly as YYYY-MM-DD and HH:MM before sending them to the database. - Preventing Double Messages: If a background check runs twice, you don't want a patient getting the exact same reminder text twice. We
used DynamoDB conditional checks so that once a reminder is claimed, no duplicate text can be sent. - Getting the Tone Right: We spent time refining the AI's prompts so its text messages sounded polite, clear, and reassuring never robotic
or cold.
──────
8. What's Next: Amazon EventBridge
Right now, a simple timer wakes up our agent to do its checks. Our next step is using Amazon EventBridge Scheduler.
With EventBridge, every time a patient books an appointment, AWS will automatically set up exact reminders: one for 24 hours before, one
for 2 hours before, and a follow-up right after. This means zero unnecessary background checks and even lower AWS costs.
──────
for 2 hours before, and a follow-up right after. This means zero unnecessary background checks and even lower AWS costs.
──────
9. Wrapping Up: AI That Truly Helps People
The AWS Agents for Humans Hackathon was all about building technology that solves everyday problems for real people.
With CareForMe, busy front-desk workers can save hours of repetitive busywork each week. Patients get easy text reminders they can answer
in two seconds. And doctors can rest easy knowing that the AI handles the admin, while humans stay in full control of patient care.
in two seconds. And doctors can rest easy knowing that the AI handles the admin, while humans stay in full control of patient care.
By combining the Strands Agents SDK with Amazon Bedrock, DynamoDB, and Cognito, we proved that AI doesn't need to replace peopleit can
simply give them their time back.
──────
simply give them their time back.
──────
Links & Project Info
• GitHub Repository: https://github.com/bellobambo/CareForMe
• Video Demo: https://www.youtube.com/watch?v=KKk44cAat4E
• Devpost Submission: CareForMe
• Built for: AWS Agents for Humans Hackathon (Professional Agents Track)
• Video Demo: https://www.youtube.com/watch?v=KKk44cAat4E
• Devpost Submission: CareForMe
• Built for: AWS Agents for Humans Hackathon (Professional Agents Track)
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article