Problem 6
Explain a symptom in plain language and connect the person to care — without diagnosing or prescribing.
When people feel sick, they often search their symptoms online, which frequently leads to unnecessary panic (e.g., WebMD telling them a cough is a rare disease). On the flip side, navigating the actual healthcare system to find the right specialist and book an appointment is highly confusing.
Standard chatbots are either too restrictive (forcing users through rigid “Yes/No” menus) or too dangerous (hallucinating medical diagnoses).
Relying on rigid menus or unconstrained chatbots can result in:
The challenge is to build an empathetic agent that operates strictly within safe boundaries — translating medical jargon, understanding symptoms contextually, and connecting the user to care.
This problem is relevant to organizations and teams including:
Build a privacy-first, empathetic AI agent that evaluates user symptoms, explains complex medical reports in simple terms, and autonomously navigates a mock healthcare system to book appointments.
The agent should maintain a calm, reassuring tone, strictly avoid diagnosing or prescribing, and convert complex medical data into plain, easy-to-understand language.
Empathetic Communication
Maintain a calm, professional, and reassuring tone.
Safety & Boundaries
Strictly refuse to officially “diagnose” or prescribe medication (include necessary disclaimers).
Jargon Translation
Convert complex medical data into 8th-grade reading level text.
The following levels are intended as guidance for increasing complexity. Participants may choose one or more levels based on their approach.
Level 1: Basic Agency A RAG-based agent loaded with verified medical documents. It listens to a user's symptoms, suggests 3 possible common ailments, and explicitly advises them to see a doctor.
Level 2: Tool Integration The agent identifies that the user likely needs a Dermatologist. It queries a mock “Hospital API” to find the nearest available Dermatologist and books an appointment for the user.
Level 3: Multi-Agent Collaboration An “Intake Agent” chats with the patient to gather a medical history. It then passes a highly structured, medically accurate summary to a “Doctor-Facing Agent,” saving the physician 15 minutes of manual chart-reading.
Level 4: Advanced / Multimodal The user uploads a PDF of a recent blood test (e.g., CBC panel). The agent uses OCR to read the results, identifies that the user's Iron levels are low, explains what that means without medical jargon, and suggests dietary changes.
A safe, conversational health assistant that bridges the gap between patient confusion and medical booking.
Participants are free to decide:
How lab reports or medical documents are ingested and explained (if implemented)
How privacy and data handling are addressed
Teams may use real, synthetic, or publicly available medical reference data where appropriate.
1. Meaningful AI Usage
The solution should go beyond a rigid Yes/No symptom menu and demonstrate real contextual understanding of symptoms.
2. Safety Boundaries Are Mandatory
The agent must never present output as a diagnosis or prescription, and must consistently include appropriate disclaimers.
3. Grounding in Verified Sources
Symptom-related suggestions should be tied to a verified medical knowledge base rather than unconstrained model generation.
4. Jargon Translation Quality
Medical terminology and lab results must be explained in plain, accessible language.
5. Machine-Usable Output
Intake summaries and booking requests should be structured so they can be consumed by a provider-facing system.
6. Privacy Awareness
The solution should handle patient information responsibly and avoid unnecessary data exposure.
7. Innovation Freedom
Participants are not restricted to a specific model, RAG corpus, or booking system.
What It Measures
Consistency in avoiding diagnosis/prescription and including disclaimers
Symptom Reasoning
Quality
Relevance and accuracy of suggested possible ailments
Jargon Translation Quality Clarity of medical explanations at an accessible reading level
Correctness of specialist matching and appointment booking
Usefulness and accuracy of the structured handoff to a provider-facing view
How calm, professional, and reassuring the interaction feels
Whether AI provides value beyond a static Yes/No symptom checker
Architecture, implementation, and robustness
Effective use of models, computation, cost, and latency
Ability to clearly demonstrate the working solution
Participants are free to choose their own technical approach. Solutions may use any relevant combination of:
The listed technologies and approaches are examples only and should not limit innovation.
Participants are encouraged to introduce new capabilities, data sources, or downstream applications that improve the value of the solution.
Given a patient's symptoms or medical documents, safely and empathetically help them understand what might be happening, in plain language, and connect them to the right care — without ever pretending to be a doctor.