Biomedical Engineer | Deep Learning & AI Specialist | LLM Enthusiast

Turning medical data into intelligent diagnostics — one model at a time.

I am a Biomedical Engineer passionaye for building AI systems that save lives. My work bridges clinical medicine and cutting-edge machine learning — from segmenting cancer cells under a microscope to deploying LLMs for medical text analysis.

I specialize in Deep Learning, Medical Image Processing, and Multimodal AI, with hands-on experience in CNNs, Transformers, RNNs, YOLO, UNet, and now Large Language Models (LLMs). I’ve built models for tumor detection, cell classification, Bitcoin forecasting, and even automated CAPTCHA solving — always with an eye toward real-world impact.

Currently exploring how LLMs can assist in clinical documentation, radiology report generation, and patient triage — because healthcare deserves intelligent automation.

Tools I use daily to turn medical data into actionable insights.

Used in: UNet segmentation of tumor cells, CBCT tooth segmentation, blood cell tracking, contrast enhancement

Used in: CNN classification of MCF-7/MDA-MB-231 cells, 3D CNN video analysis, CIFAR-10 modifications, skin cancer detection

Used in: Statistical analysis of cell movement, model performance comparison, visualization of CBCT results, Bitcoin price forecasting

Used in: Spam email detection, LLM fine-tuning on Persian medical notes, scraping PubMed abstracts, clinical report summarization

Used in: Blood cell tracking (YOLOv8), cell boundary detection, data augmentation for low-sample medical images

Used in: Training loop monitoring, model serialization, config management, interactive experimentation

• Fine-tuned Llama 3 and Mistral on clinical notes for symptom extraction and diagnosis suggestion.
• Built a RAG (Retrieval-Augmented Generation) pipeline using LlamaIndex + FAISS to answer medical questions from PubMed and hospital records.
• Goal: Reduce clinician burnout by automating documentation and triage.

• Trained a BERT-based classifier on SMS and email datasets (SMS Spam Collection, Enron Email).
• Achieved 98.2% accuracy using Hugging Face transformers + scikit-learn.
• Deployed as a Flask API for real-time filtering.

• Built a multi-digit CAPTCHA solver using CNN + CTC Loss on synthetic CAPTCHA images.
• Used Tesseract OCR and CRNN (Convolutional Recurrent Neural Network) for robust character recognition.
• Performance: >95% accuracy on distorted, noisy CAPTCHAs.

• Developed a CNN pipeline to classify and number 32 human teeth in Cone-Beam CT images.
• Combined UNet segmentation with template matching to label teeth (e.g., #11, #36) automatically.
• Reduces manual annotation time by 80% — ready for dental AI clinics.

• Classified 10,000+ dermoscopic images of skin lesions (melanoma, nevus, basal cell carcinoma) using EfficientNet-B4.
• Achieved 94.1% accuracy with data augmentation and transfer learning.
• Integrated Grad-CAM to visualize decision regions — critical for clinical trust.

• Built a UNet architecture to segment MCF-7 and MDA-MB-231 breast cancer cells from microscopy images.
• Trained a CNN classifier to distinguish cancer types with 92.3% accuracy.
• Published in To Be Submitted journals.

• Analyzed time-lapse videos of cancer cells using 3D CNN + LSTM to predict metastatic behavior.
• Enabled early detection of aggressive phenotypes — key for personalized therapy.

• Applied YOLOv8 for real-time detection and SORT for multi-object tracking of RBCs/WBCs.
• Analyzed cell velocity and interaction patterns — useful for hematology research.

• Trained RNNs on Yahoo Finance data to forecast Bitcoin prices with 87% directional accuracy.
• Explored attention mechanisms and ensemble forecasting.

• Enhanced ResNet and DenseNet architectures to improve accuracy from 90% → 94.5%.
• Implemented label smoothing, mixup, and dropout optimization.

• Implemented Active Contour (Snake) algorithm for precise dental segmentation.
• Outperformed thresholding methods in edge-preserving accuracy.

📧 aziziali.9473@gmail.com
🔗 LinkedIn
🏆 Kaggle
🐙 GitHub

• Multimodal AI: Combining images + text for radiology diagnosis
• Edge AI: Deploying models on low-power medical devices
• Persian NLP: Building LLMs trained on Persian medical literature
• AI Ethics in Healthcare: Bias detection in diagnostic models

I’m actively seeking opportunities to:

• Work on AI for healthcare startups or research labs
• Contribute to open-source medical AI tools
• Join a team building LLMs for clinical decision support

Let’s build AI that doesn’t just predict — but heals. 🤝

📚 Reading Books • ⚽ Watching Soccer • 🎵 Listening to Music • 🎮 Playing Video Games • 🎬 Watching Movies