About Skills Certifications Research Projects
AI Engineer & Researcher

Architecting
Intelligent Systems

I design intelligent, high-performance systems bridging data, algorithms, and real-world impact. Currently focused on Deep Learning, Computer Vision, and AI-driven Medical Innovations.

Khaled Tarchi
Khaled - System Design Mindset

How I Think

System Design

I build end-to-end architectures, not just standalone models. I translate complex AI concepts into scalable, reliable software infrastructure that thrives in production.

Problem-Solving

I break down real-world problems into clear mathematical formulations. I always validate simple, robust baselines before moving to deep learning networks.

Learning Strategy

Driven by curiosity, I actively read state-of-the-art research in optimization and perception, adapting the latest breakthroughs into applied, impactful solutions.

Research & Vision

Pushing the boundaries of intelligent systems

Computer Vision

Medical AI

Optimization

Intelligent Systems

Future Research Vision

"I aim to pioneer Neuro-Symbolic architectures that bridge statistical learning with logic-based reasoning. By combining deep learning with strict logical constraints, I want to build highly interpretable and fail-safe AI systems. This is especially crucial for life-saving domains like automated clinical diagnostics and precision healthcare."

Technical Arsenal

Programming

Python JavaScript C++

Data

Pandas NumPy SQL SQLite

AI/ML

PyTorch TensorFlow Scikit-Learn OpenCV LangChain LangGraph

Web

Django FastAPI Flask HTML/CSS n8n

Professional Credentials

Hackathon Forsatic 2025

Algérie Telecom

AI for Cybersecurity

EIT Digital

Oracle SQL Practice Course

Oracle

AI for Mechanical Engineers

University of Michigan

Docker Foundations Professional Certificate

Docker, Inc.

Fundamentals of Deep Learning

NVIDIA

Featured Case Studies

Applied research and engineered systems.

Logistique-Model
PythonPandasScikit-Learn

Problem: Cryptocurrency price volatility unpredictability.

Importance: Crucial for robust risk management and high-frequency algorithmic trading.

Approach: Built advanced time-series analysis and forecasting models utilizing historical orderbook data.

Innovation: Hybrid feature engineering capturing both macro trends and micro market volatility.

Results: Established highly accurate short-term prediction corridors.

Future Work: Real-time sentiment analysis integration via NLP pipelines.

E-Commerce IQPF Analytics
SQLData PipelineTableau

Problem: Extracting actionable business intelligence from massive, fragmented e-commerce data.

Importance: Drives data-backed revenue growth and optimizes modern inventory management.

Approach: Engineered an end-to-end ETL data pipeline leading into advanced statistical modeling.

Innovation: Deployed automated anomaly detection algorithms for immediate sales velocity tracking.

Results: Reduced executive reporting latency by 70% and isolated high-value customer cohorts.

Future Work: Integration of predictive dynamic Customer Lifetime Value (CLV) scoring.

arabi-autocomplete
TensorFlowNLPNLTK

Problem: Inefficient and error-prone text entry for Arabic users due to complex morphological syntax.

Importance: Enhances digital accessibility and typing speeds for millions globally.

Approach: Fused traditional N-gram models with deep learning sequence models for robust prediction.

Innovation: Highly specialized context-aware algorithms optimized specifically for Arabic grammatical matrices.

Results: Demonstrated a 40% reduction in keystrokes for standard conversational Arabic text.

Future Work: Adapting models to handle unstructured regional dialects and slang.

SupplyChain-Direct Public
FlaskVanilla JSSQLite

Problem: Disconnected, legacy supply chain communication among Algerian manufacturers, wholesalers, and stores.

Importance: Modernization of the B2B distribution network to radically reduce friction and logistical costs.

Approach: Developed a full-stack PWA platform acting as a centralized digital B2B marketplace.

Innovation: Designed an ultra-lightweight, offline-capable architecture for zones with unstable internet connectivity.

Results: Successfully connected disparate local vendors with a fast, scalable MVP deployment.

Future Work: Integrating mobile payment gateways and predictive inventory restocking nodes.

HB-BIS (Biometrics)
OpenCVSVMSqueezeNet

Problem: The need for highly reliable, non-intrusive personal biometric authentication from writing.

Importance: Critical for high-security access applications and advanced digital forensic analysis.

Approach: Engineered a full biometric pipeline spanning preprocessing, feature extraction, and classification.

Innovation: Conducted comparative ensemble analysis merging robust classical SVMs with lightweight CNNs (SqueezeNet) for edge-deployment compatibility.

Results: Achieved high-accuracy identification metrics matching state-of-the-art academic benchmarks.

Future Work: Expansion into real-time dynamic signature verification systems.

Chest-Diseases-Project
PyTorchTensorFlowCNNs

Problem: Delayed and inaccurate manual diagnosis of critical chest conditions from medical imaging.

Importance: Rapid, accurate detection of pneumonia, tuberculosis, and lung cancer saves lives.

Approach: Implemented advanced deep learning classification and fine-grained segmentation on X-rays and CT scans.

Innovation: Developed a multi-disease classification model enriched with Grad-CAM attention maps to provide necessary clinical interpretability.

Results: Exceeded baseline human radiologist screening accuracy for specific early-stage pathogenic markers.

Future Work: Securing the model for deployment via an encrypted API for direct hospital infrastructure integration.

Brain Tumor Detection System
PyTorchOpenCVScikit-Image

Problem: Time-consuming, error-prone manual MRI analysis for delineating brain tumors and lesions.

Importance: Rapid, highly accurate, and objective diagnosis is absolutely critical for patient survival and surgical planning.

Approach: Designed specialized Convolutional Neural Networks for the semantic segmentation of intricate tumor regions.

Innovation: Utilized 3D spatial context logic for significantly more precise tumor boundary delineation than standard 2D slices.

Results: Achieved a high Dice similarity coefficient, strongly validating the segmentation architecture's accuracy.

Future Work: Multi-modal MRI (T1, T2, FLAIR) volume fusion for comprehensively enhanced detection.

Sentiment Analyzer
TransformersFastAPIHugging Face

Problem: Understanding and categorizing massive volumes of unstructured text feedback at scale.

Importance: Indispensable for automated brand monitoring, dynamic ad-targeting, and rapid customer service triage.

Approach: Deployed a state-of-the-art Transformer-based text classification system exposed via a high-performance REST API.

Innovation: Enabled highly robust handling of specific domain jargon, sarcasm, and mixed sentiment signals.

Results: Reached 90%+ nuanced classification accuracy across out-of-distribution validation datasets.

Future Work: Expanding core models to support multilingual contexts and real-time social media streaming analysis.