HackNation · Women's Hormonal Health

Team Artificially Unintelligent

For a submission to the 6th Hack-Nation Global AI Hackathon, we built this application. A clinical-support prototype for women's hormonal health that combines multiple specialised models instead of trusting a single predictor — then connects the result to a shared patient record workflow, role-based access, and a safety-aware health chatbot.

Watch the demo ↓ View source on GitHub ↗

3 model perspectives Generic thyroid · women-only PCOS · clinical rules

2 conditions PCOS + thyroid dysfunction

2 user roles Patient + clinician workflows

The idea

Don't throw away imperfect models. Make their limitations useful.

The project challenges the idea that one dataset or one model should carry the full decision. Instead, different sources of evidence contribute to a combined output, while the application keeps the final interpretation in the hands of a clinician.

Dataset bias

Generic clinical models may inherit mixed-sex or male-skewed training distributions that do not represent women's hormonal health equally well.

Narrow specialisation

Women-only models can be more targeted, but a narrow cohort may not capture every relevant signal or comorbidity on its own.

Fragmented records

Patient history is often distributed across separate providers. The prototype explores a shared record layer with controlled patient and doctor permissions.

System architecture

Multiple signals in. One clinically focused result out.

Uploaded lab and screening information is transformed into structured features and evaluated through complementary model paths. Their outputs are combined into a final result rather than treating any individual model as absolute ground truth.

Input

Patient workup

Lab and screening PDF
Hormonal / thyroid values
Clinical and ultrasound features

→

Inference

Complementary models

Mixed-sex thyroid model
Women-only PCOS model
Transparent clinical-rule engine

→

Output

Decision support

Condition-oriented confidence
Reasoning / supporting signals
Explicit missing-data handling

Model A

Thyroid cohort

Uses mixed-sex thyroid data containing TSH, T3, T4, T4U, FTI and patient context.

Model B

PCOS cohort

Uses women-only hormonal, menstrual, symptom and ultrasound/follicle information.

Model C

Clinical criteria

Encodes interpretable diagnostic rules so the pipeline is not dependent only on learned statistical patterns.

Interface gallery

The application, from patient records to study analysis.

A walkthrough of the patient and clinical workflows, from secure access and health records to AI-assisted analysis and human escalation. Use the arrows or dots to explore the interface.

Application login page for patient and staff access

Unified Login A shared authentication entry point that routes users into patient or clinical-staff workflows.

Patient homepage with health records and AI chatbot access

Patient Dashboard A patient-facing hub for longitudinal health records, AI-assisted guidance and clinical communication.

Staff homepage with patient records, study analysis and patient help requests

Clinical Staff Dashboard Centralizes patient records, AI-assisted study analysis and pending patient support requests.

Patient view of personal medical records

Patient Health Records Provides read-only access to medical history across participating care providers in one interface.

AI Health Assistant Answers general health questions while maintaining clear boundaries around diagnosis and clinical advice.

AI chatbot detecting a potentially serious health scenario

Safety Escalation Detects potentially serious health concerns and redirects the patient toward appropriate human or emergency support.

Patient interface for asking a doctor a question

Ask a Doctor Escalates patient questions to clinical staff when AI guidance is insufficient or human review is preferred.

Staff interface for searching and reviewing patient records

Clinical Record Search Allows authorized staff to retrieve patient histories and review records through a unified clinical interface.

Staff AI-assisted health study analysis interface

AI Study Analysis Processes uploaded health results through the ensemble pipeline to support evidence-grounded clinical review.

Staff interface for viewing and answering patient questions

Patient Help Queue Surfaces unanswered patient questions so clinical staff can review context and respond directly.

Team brainstorming and project development discussion

Behind the Build From architecture sketches to rapid iteration: collaborative brainstorming behind the system design and prototype.

Demo video

See the submission video

This video includes introduction of the team and a demo of the application showcasing patient record access, chatbot escalation, clinician upload permissions, the doctor help queue, and the study-analysis pipeline.

Application layer

AI is only one part of the product.

The prototype wraps the model inside a role-aware healthcare workflow, separating what patients can do from what clinicians can do and providing a human escalation path for chatbot interactions.

◫

Shared health records

Patient history from participating providers is presented through a common record view, reducing the need to navigate isolated hospital portals.

⌁

Role-based permissions

Patients can read and download their records, while clinicians additionally receive upload access for adding new reports and screening results.

✦

Safety-aware chatbot

The health assistant is positioned as an informational interface rather than a diagnostic tool. Serious queries can be summarised and escalated for doctor review.

∿

Human-in-the-loop analysis

The model surfaces evidence for further investigation; the clinician remains responsible for interpreting the result and making the medical decision.

Datasets

Two cohorts with deliberately different perspectives.

The project uses a women-only PCOS cohort alongside a mixed-sex thyroid cohort, allowing the system to explore how specialised and generic data sources can contribute differently.

PCOS Cohort

Prasoon Kottarathil · Polycystic Ovary Syndrome (PCOS) Dataset

541 women Hormone panels Ultrasound Clinical symptoms

Clinician-confirmed PCOS labels
Menstrual-cycle history
Follicle and ultrasound measurements

Thyroid Cohort

UCI Machine Learning Repository · Thyroid Disease / Hypothyroid Dataset

~3,160 patients 69% female 31% male Thyroid labs

TSH, T3, T4, T4U and FTI
Age and sex features
Pregnancy and acute-illness indicators

Prototype context: this project was created as an AI hackathon prototype and is not a medical diagnostic device. Its design intentionally keeps diagnosis and clinical judgment with healthcare professionals.

Implementation

Built as an end-to-end ML + web application prototype.

The documented implementation centres on a Flask application, reproducible dataset processing, machine-learning inference, persistent patient records, and an AI-assisted patient interaction layer.

Python Flask pandas scikit-learn Machine Learning Ensemble Modelling Clinical Rule Engine PDF / Lab Extraction Role-Based Access AI Chatbot

Explore the build

Models should expose uncertainty — not hide it.

Artificially Unintelligent explores a healthcare AI workflow where specialised models, clinical logic, missing-data awareness and human review work together instead of relying on a single opaque prediction.

Open GitHub repository ↗