The problem

Healthcare AI faces a fundamental limitation: lack of high-quality, real-world conversational data.

Most existing systems are trained on:

• Synthetic datasets
• Structured clinical records
• Limited or highly controlled data sources

However, real patients do not describe symptoms in structured formats. They communicate in:

• Informal language
• Incomplete descriptions
• Ambiguous or subjective terms

At the same time, strict privacy regulations severely limit access to real clinical conversations.

This creates a critical gap:

• AI systems struggle to interpret real patient language
• Triage tools lack reliability in real-world scenarios
• Healthcare providers face inefficiencies in early-stage patient assessment

Despite the availability of public medical conversations (forums, archives), this data remains:

• Unstructured
• Noisy
• Not clinically labeled
• Not safe for direct use due to privacy concerns

The project goals

This project proposes building Medi-Triage Core, a medical intelligence system designed to transform public conversational data into a structured, privacy-safe foundation for AI-driven triage.

The solution focuses on creating a high-fidelity, de-identified clinical dialogue dataset and leveraging it to train specialized NLP models.

Key components include:

• Aggregating conversational data from public medical sources (Reddit, NHS, CDC archives)
• Building a de-identification pipeline to remove all personal information
• Designing a clinical annotation schema for symptoms, intent, and urgency
• Developing NLP models for symptom detection and triage classification
• Creating a triage chatbot prototype for real-time interaction
• Delivering a monitoring dashboard for system outputs

As part of this challenge, the system must demonstrate the ability to:

• Convert raw medical conversations into structured, labeled datasets
• Ensure strict privacy compliance through robust anonymization
• Accurately identify symptoms and medical entities
• Classify urgency levels (Low, Medium, High)
• Handle real-world, ambiguous patient language
• Simulate triage conversations through a chatbot interface
• Highlight uncertainty and limitations in medical interpretation 

Impact of the Problem

If successful, Medi-Triage Core can directly improve how patients access care and how healthcare systems manage demand.

Patients & General Public

• Faster initial assessment of symptoms
• Better guidance on when to seek urgent care
• Reduced uncertainty in early-stage health concerns
• Improved access to basic triage support

Healthcare Providers

• Reduced overload in emergency and primary care services
• Better prioritization of high-risk cases
• More efficient intake and triage workflows
• Support for telehealth and remote care systems

Public Health Systems

• Scalable triage support during high-demand periods (e.g., outbreaks)
• Improved allocation of medical resources
• Early identification of emerging health trends from conversational data

Healthcare Innovation

• Creation of privacy-compliant, high-quality medical datasets
• Foundation for next-generation clinical NLP systems
• Acceleration of safe AI adoption in healthcare

Real-World Impact

• Shorter wait times for critical patients
• Reduced strain on the healthcare infrastructure
• Safer, more accessible first-line medical guidance at scale

Timeline

1

The Foundation (Weeks 1-2). Establishing the technical architecture and aggregating data from trusted public health archives and forums.

2

Privacy & Expert Labeling (Weeks 3-4). Transforming raw dialogue into a privacy-compliant dataset, verified by medical experts.

3

Sprint 3: Intelligence Training (Weeks 5-6). Developing the core NLP models to detect symptoms and understand clinical intent.

4

Sprint 4: Integration & Delivery (Weeks 7-8). Building the functional chatbot interface and final medical monitoring dashboard.

 

**More details will be shared with the designated team.