Leddycia/STD-MedGuide

Model Card for Model ID

STD-MedGuide is an AI model designed to answer questions about sexually transmitted diseases (STDs/MSTs) in clear, natural language.

You give it a question (e.g., “What are the common symptoms of chlamydia?”).

The model processes the question and generates a medically relevant answer, based on knowledge it learned from medical QA datasets.

It is intended for education and awareness, not as a substitute for professional medical advice.

Model Details

🛠️ Model Details – STD-MedGuide

Model Name: STD-MedGuide

Base Architecture: FLAN-T5 (small) — a text-to-text transformer model designed for instruction-following tasks.

Task: Question Answering (QA) focused on sexually transmitted diseases (STDs/MSTs).

Dataset Used:

CareQA (open-ended medical QA dataset)

Medical-QA (open-ended medical QA dataset)

Both datasets adapted for French MST terminology and general STD questions.

Training Environment: Google Colab using Hugging Face transformers library.

Training Setup:

Epochs: 2 planned (beta version checkpoint at step ~171)

Batch Size: 4–8 (depending on GPU/CPU availability)

Learning Rate: 5e-5

Checkpointing: every 100 steps (to save progress and allow beta sharing)

Device: GPU if available (Colab runtime), otherwise CPU

Loss Performance:

Initial training loss ~21

Reduced to ~0.5 in early steps (checkpoint 171)

Tokenizer: AutoTokenizer from Hugging Face, compatible with FLAN-T5 small.

Output Type: Text generation — given a question, generates an answer in natural language.

Intended Users: Students, educators, awareness programs, and general audiences seeking information about STDs.

Limitations: Not suitable for clinical diagnosis; probabilistic output; may reflect dataset biases.

Model Description

STD-MedGuide is a specialized question-answering model designed to provide reliable and accessible information about sexually transmitted diseases (STDs/MSTs). Built on top of the FLAN-T5 architecture, it has been fine-tuned using medical QA datasets to understand and respond to natural language questions related to sexual health.

The model’s primary function is to take a user’s question — for example, “Quels sont les symptômes les plus courants des MST ?” or “How can HIV be prevented?” — and generate a concise, contextually relevant answer. By focusing specifically on STD-related queries, STD-MedGuide can assist students, educators, and awareness programs in spreading accurate information about prevention, symptoms, transmission methods, and general health education.

While it produces medically informed responses, STD-MedGuide is not a diagnostic or clinical tool. Instead, it is intended as an educational resource, making medical knowledge more approachable and helping raise awareness about sensitive health topics. Its potential applications include supporting classroom learning, contributing to health campaigns, and serving as an interactive guide for individuals seeking general knowledge about STDs.

• Developed by: [Saint-Vil Angie-Reyna Leddycia]

Uses

🏷️ Uses / Applications of STD-MedGuide

Educational Tool

Helps students, teachers, and healthcare trainees learn about sexually transmitted diseases.

Can be integrated into classroom exercises, quizzes, or self-study activities.

Awareness and Prevention Campaigns

Provides clear, understandable answers for public health awareness.

Useful for NGOs, health organizations, or school programs promoting sexual health.

Interactive Question-Answering

Can serve as a chatbot or interactive assistant to answer general STD questions.

Supports engagement in workshops, seminars, or online learning platforms.

Reference for Research or Reports

Can generate concise explanations of STD symptoms, transmission, and prevention for reports, presentations, or educational materials.

Language Adaptation

Handles French terminology (MST) and English (STD), making it suitable for bilingual or international educational settings.

Note: While STD-MedGuide is a powerful educational tool, it is not a substitute for professional medical advice or clinical diagnosis.

Bias, Risks, and Limitations

⚠️ Bias, Risks, and Limitations

Bias in Training Data

STD-MedGuide is fine-tuned on datasets like CareQA and Medical-QA, which may reflect biases in medical literature, language, or cultural context.

Certain diseases, symptoms, or populations may be over- or under-represented, affecting the accuracy or completeness of answers.

Medical Accuracy and Reliability

The model generates responses based on patterns in the training data, not on real-time clinical knowledge.

Answers may occasionally be incomplete, outdated, or partially incorrect, so it cannot replace professional medical advice.

Language and Context Limitations

While the model handles French (MST terminology) and English reasonably well, complex questions or nuanced phrasing may lead to misinterpretation.

It may not fully understand ambiguous questions or rare medical scenarios.

Ethical and Safety Risks

Users could misinterpret answers as formal medical guidance.

There is a risk of over-reliance, especially by individuals seeking medical help without consulting healthcare professionals.

Technical Limitations

The current beta version is trained on a limited number of steps (~171), meaning it has not seen the full dataset and could improve with further training.

Responses are generated probabilistically, so the model may produce different answers to the same question on repeated queries.

Conclusion: STD-MedGuide is best used as an educational and awareness tool, to support learning and information dissemination about STDs, but not for diagnosis or clinical decision-making.

[More Information Needed]

Recommendations

💡 Recommendations

Further Training

Continue training on the full dataset to improve accuracy and coverage of STD-related questions.

Consider incorporating additional French-language medical datasets to improve bilingual performance.

Regular Updates

Periodically update the model with new medical knowledge and guidelines to ensure answers stay current.

Evaluation and Validation

Evaluate on a larger, diverse validation set to measure performance and reduce bias.

Consider human-in-the-loop review to check medical accuracy of generated answers.

Responsible Use

Use the model primarily for education and awareness, not for clinical decision-making.

Always include disclaimers when deploying the model to inform users of its limitations.

User Interface & Accessibility

Integrate into chatbots, educational platforms, or mobile apps for easier access.

Consider adding multilingual support beyond French and English.

Bias Mitigation

Monitor for biases in answers (e.g., population, gender, age) and refine the training dataset accordingly.

Encourage feedback from users to improve answer fairness and relevance.