Sebia Kods

Machine Learning Engineer — Deep Learning · Reinforcement Learning · Healthcare AI

Building applied ML systems from raw data to deployed models, with a research focus on offline reinforcement learning for clinical decision-making.

About

M.Sc. graduate in Fondements et Ingénierie de l'Information et de l'Image (F3I), Université Ferhat Abbas Sétif (2024–2026), following a B.Sc. in Computer Science from the same institution.

My work sits at the intersection of reinforcement learning and healthcare — using clinical time-series data to build models that support real decisions, not just benchmark scores. I care about the full pipeline: data engineering, model development, evaluation, and getting something into a usable form.

Currently seeking entry-level roles in Machine Learning Engineering, AI Engineering, or Applied Research.

Technical Skills

Category	Tools
Languages	Python · SQL
ML / DL Frameworks	PyTorch · TensorFlow · scikit-learn
Data	Pandas · NumPy
AI / NLP	LangChain · Hugging Face Transformers
Infrastructure	Git · REST APIs · Docker (learning)

Research & Projects

Offline Reinforcement Learning for Insulin Dosing

Learned a personalized insulin dosing policy from real ICU time-series data (MIMIC-III), using offline RL to avoid the risk of online exploration in a clinical setting.

• Built the full data pipeline: cohort extraction, state representation, and reward shaping from raw ICU records
• Framed dosing as a sequential decision problem rather than a single-step prediction
• Stack: Python · PyTorch · Offline RL
• Repository →

Forest Fire Risk Prediction — Algeria

Multi-model system comparing Random Forest, SVM, Logistic Regression, and neural approaches for fire risk classification, backed by a real-time weather API and interactive risk maps.

• Compared model families head-to-head rather than committing to one algorithm upfront
• Shipped as a usable tool, not just a notebook — real-time API + map interface
• Stack: Python · scikit-learn · Flask
• Repository →

Academic Paper Search Engine

Information retrieval system built from scratch over arXiv papers — inverted indexing, Boolean (AND/OR) search, phrase and proximity search, and TF-IDF ranking.

• Implemented core IR algorithms directly rather than relying on a search library
• Stack: Python · NLP · Information Retrieval
• Repository →

GitHub Activity

Currently Exploring

• Fine-tuning open-source LLMs for domain-specific clinical applications
• Extending the offline RL work toward a full decision-support prototype
• Deploying trained models as production-ready APIs

---

_{Open to connecting with people working in AI, ML, or healthcare tech.}