Deep Learning Researcher & AI Systems Engineer ENSIA - National Higher School of Artificial Intelligence
Algiers, Algeria

Architecting research-grade deep learning models, distributed optimization algorithms, and scalable AI backend systems. Currently focused on bounded gradient quantization, epistemic uncertainty estimation, and deploying production-ready ML infrastructure. I bridge the gap between rigorous mathematical theory and high-throughput PyTorch implementations.

class SamirGuenchi:
    def __init__(self):
        self.role = "Deep Learning Researcher & AI Systems Engineer"
        self.institution = "ENSIA (Class of 2027)"
        self.location = "Algiers, Algeria"
        self.id = 194
        self.focus = [
            "Distributed Optimization (DDP)",
            "Epistemic Uncertainty Estimation", 
            "Multi-Agent LLM Orchestration",
            "High-Throughput Backend Architecture"
        ]
        
    def current_work(self):
        return {
            "research": ["MoQAdam Optimizer", "QuantumPath Neural Networks (QPNN-v2)"],
            "applied_ai": "SPIRAL-RAG v3 & Clinical Sequence Modeling (SleepStagerNetV4)",
            "systems": "AI-Automated E-commerce Ecosystems (Node.js/TypeScript)",
            "leadership": "Coach for the Algerian Olympiad in Informatics (AOI)"
        }
    
    def philosophy(self):
        return "Theories that scale > Math that stays on paper."

• Arabic Natural Language Processing
• Retrieval-Augmented Generation (RAG) Systems
• Competitive Programming & Algorithm Design
• Machine Learning Security
• Medical AI & Healthcare Applications

A novel distributed PyTorch optimizer replacing AdamW for bandwidth-bottlenecked Multi-GPU (DDP) training.

• Formulated Variance-Normalized Quantization and SNR-Gated Residual Feedback.
• Engineered Decoupled Moment Updates to preserve unbiased variance estimation, reducing state memory footprint to 2.5P.

A quantum-inspired neural network architecture for principled epistemic uncertainty estimation and OOD detection.

• Engineered Complex Path Representations (CPR) and Orthogonal Interference Mixing.
• Architected to reduce parameter count by ~70% (12.6M vs. 44.8M) compared to deep ensembles while targeting improved ECE calibration metrics.

GPU-accelerated clinical sequence modeling pipeline aiming to approach expert inter-rater agreement (targeting Cohen's κ ≥ 0.81).

• Designed a Differentiable STFT + Cross-Attention fusion module to gate spectral evidence against time-domain embeddings.
• Integrates Supervised Contrastive Loss (SupCon) and FlexMatch adaptive pseudo-labeling.

An experimental RAG pipeline processing 8,600+ Algerian legal documents across 4 languages with Semantic Authority Classification.

• Implemented a Multi-Agent Legal Debate reasoning pipeline using Llama-3.3 and Gemini 2.0 Flash to synthesize complex regulatory interpretations.

Cross-platform AI ecosystem (React, React Native, FastAPI) for multilingual drug addiction awareness.

• Architected an LLM orchestrator utilizing MarBERT for intent classification and OOD detection.
• Deployed a scientific RAG (ChromaDB) and an empathetic support model (Qwen2.5-7B + LoRA/PEFT).

Architected and deployed highly scalable platforms currently operating in production for active retail businesses.

• Automation Suite: Engineered NLP sentiment analysis for product trends, LLM customer auto-replies, and 1-click omnichannel social media publishing.
• Mahal Dahaoui: Live marketplace (MongoDB/React) with real-time state synchronization via Socket.IO.
• Djellaba El Basma: 3-tier system (React/Node/React Native) scaling to 170+ deployments.

• Coach, Algerian Olympiad in Informatics (AOI): Mentoring elite national competitors in advanced data structures, graph theory, and dynamic programming (2025).
• Competitive Programming: 300+ advanced algorithmic problems solved across Codeforces and LeetCode.
• Hackathons & Datathons: * 2023 Data Science Datathon (Predictive modeling for Yassir & Société Générale).
• ETCODE 2023 & Mobile Ideathon.
• Covid-19 Rapid Prototyping Workshop (ENSIA & IPSIL).

Institution: ENSIA Program: Computer Science & AI Engineering
Location: Algeria
Focus Areas: NLP, Machine Learning, Algorithm Design

Additional Role: National Programming Olympiad Coach

Seeking Research Internships in well-resourced ML labs to access the computational clusters required to benchmark my ongoing algorithmic research (MoQAdam & QPNN-v2) and contribute to foundational ML systems. Research:

• Arabic RAG architectures
• Morphological tokenization systems
• Multilingual NLP infrastructure

Development:

• Production ML systems deployment
• Open-source Arabic language tools
• Security-focused ML applications

Education:

• Training national olympiad candidates
• Algorithm optimization techniques
• Competitive programming methodology

Seeking:

• Research collaborations in Arabic NLP
• ML/NLP engineering internships
• Open-source contribution opportunities
• Hackathons and ML competitions

Professional:

• LinkedIn: linkedin.com/in/guenchi-samir
• Email: samir.guenchi@ensia.edu.dz
• GitHub: github.com/Samir-Guenchi
• Phone : +213780066761.

Competitive Programming:

• Kaggle: kaggle.com/guenchisamir
• Codeforces: codeforces.com/profile/Guenchi_Samir_ia

Last Updated: MAY 2026