🚀 Launching Our AI-Powered Turkish Health Support Chatbot! 🇹🇷💬
In regions where healthcare access is limited, we built a 24/7 Turkish-language chatbot that provides users with fast, reliable answers to basic health-related questions using cutting-edge LLM and NLP technologies.
🧠 🔹 Project Overview:
Users can ask natural questions like: “I have a headache” or “I feel nauseous”, and the bot replies with possible causes and suggestions.
Designed for native Turkish speakers and optimized to improve health literacy and reduce unnecessary hospital visits.
Accessible, real-time, and developed with a focus on public benefit.
📊 🔹 Dataset Information:
Format: CSV file with 15,000 question-answer pairs.
Source: Translated from SQuAD (Stanford Question Answering Dataset).
Sample:
Q: “How can I relieve a headache?”
A: “Rest, drink plenty of water, and take painkillers if needed.”
🧹 🔹 Data Preprocessing Steps:
Text cleaning: Removed HTML tags, links, and special characters.
Normalization: Lowercasing, punctuation handling, whitespace trimming.
Tokenization: Used meta-llama/Llama-3.2-1B-Instruct tokenizer for LLM compatibility.
Libraries: transformers, datasets, os, torch, Flask.
🔧 🔹 Model Development:
Model: meta-llama/Llama-3.2-1B-Instruct – a 1B parameter Turkish-tuned LLaMA 3 model.
Architecture: Causal decoder, fine-tuned on domain-specific healthcare QA data.
Training Configuration:
Epochs: 3
Batch Size: 8
Learning Rate: 2e-5
Optimizer: AdamW
Results:
Initial: Loss = 2.78, Accuracy = 11%
Final: Loss = 0.12, Accuracy = 73%
Training loss decreased steadily, indicating strong learning performance.
🌐 🔹 Web Interface:
Built with Flask for seamless user interaction.
Users submit questions through a simple HTML interface.
Backend:
Checks if the question was asked before.
If new, the model generates and stores the answer.
Responses are returned in JSON.
“Clear Chat” button allows resetting the session.
💡 🔹 Project Impact:
✅ Promotes Turkish-language NLP applications
✅ Real-world health chatbot use-case using LLaMA 3
✅ End-to-end AI integration (data, training, deployment)
✅ Fully functional Flask web app with real-time responses
👨💻 Developer: Taqi Eddine El Mamouni
👥 Teammate: ILYASS ELMAMOUNI
🎓 Advisor: Dr. Kadir TOHMA
📅 Project Date: May 29, 2025
hashtag#AI hashtag#HealthcareAI hashtag#LLM hashtag#LLaMA3 hashtag#NLP hashtag#TurkishLanguage hashtag#DeepLearning hashtag#Chatbot hashtag#MachineLearning hashtag#Flask hashtag#OpenSource hashtag#TaqiEddineElMamouni hashtag#HealthTech hashtag#DataScience
Top comments (0)