✦ AI Engineer · ML Architect · Data Scientist
Building intelligent systems at the intersection of data and decisions
Turning raw data into scalable ML pipelines and production-grade AI. Passionate about every layer of the stack — from preprocessing to deployment.
About Me
I'm Sebia Kods, an AI Engineer and ML Architect based in El Eulma, Sétif, Algeria. I work across the full ML spectrum — from building NLP pipelines and computer vision systems to deploying real-time AI dashboards and offline reinforcement learning for clinical settings.
My work spans some public projects on GitHub, an M.Sc. in Computer Science specializing in AI, and a deep interest in making models that work reliably in production — not just in notebooks.
I care about clean code, rigorous evaluation, and systems that hold up when the data gets messy and the stakes get real.
Technical Skills
Projects
End-to-end automated ML pipeline covering feature engineering, model selection, training, and deployment. Designed for reproducibility and production readiness.
Full pipeline · Auto feature engineering · Docker deployment
Real-time data visualization platform powered by predictive ML models. FastAPI backend serves model predictions to a dynamic TypeScript frontend.
Real-time predictions · FastAPI + TypeScript
Modular toolkit for text classification and LLM fine-tuning. Built on HuggingFace Transformers and LangChain for flexible NLP experimentation and research workflows.
Modular · LLM fine-tuning · HuggingFace
Forest fire prediction system for Algeria using Random Forest with multi-model comparison — Logistic Regression, SVM, and Neural Networks. Includes a real-time API and interactive maps.
96.7% accuracy · Real-time API · Interactive maps
NLP-based classifier that distinguishes spam from legitimate SMS messages using SVM and classic text preprocessing — tokenization, TF-IDF, and feature extraction.
NLP · SVM · TF-IDF pipeline
Computer vision system for real-time facial detection and recognition using deep learning and classical CV techniques. Built with OpenCV and Python.
Real-time detection · Deep learning + CV
Full desktop application for image processing and analysis — edge detection, Gaussian filtering, noise addition, and segmentation methods including Otsu, K-Means, and Region Growing.
Desktop GUI · 6+ segmentation methods
Information retrieval pipeline collecting and indexing up to 2,000 arXiv articles. Features inverted indexing, Boolean/phrase/proximity search, and TF-IDF cosine similarity ranking.
2,000 papers indexed · Multi-mode search
Research
Designed and trained an offline RL system (BCQ & CQL) on 146K clinical transitions from MIMIC-III to personalize insulin dosing. Built a custom MDP with multi-component rewards, clinical feature engineering from raw EHR data, and evaluated learned policies across 7 axes — OPE estimators, safety metrics, clinician agreement, and subgroup analysis.
Education
Let's Connect
Research collaborations, ML engineering roles, interesting projects — if you're building something that matters, let's talk.