Final-year AI & Data Science Engineering student at ENSAM Meknès, Université Moulay Ismaïl (Morocco), graduating June 2027. Specializing in AI and Data Science applied to industrial systems.
Research Focus
Work sits at the intersection of explainable AI, multi-agent LLM systems, and autonomous perception — with a common thread of making complex, multi-stage AI pipelines transparent and auditable rather than treating them as black boxes.
Applied Practice
Alongside research, spent the last two years building and shipping production AI systems across fintech, healthcare imaging, and applied machine learning tooling.
Background
Applied ML spanning computer vision, time-series, and NLP, plus production AI systems engineering across three internships and one independent AI agent platform.
Hierarchical Explanation Graph (HEG): Causal Explainability for Hierarchical LLM Multi-Agent Pipelines
HEG introduces a causal explainability framework for hierarchical, multi-agent LLM pipelines — systems where multiple orchestrators, planners, and executors interact across layers, making failure attribution notoriously difficult. A dual-orchestrator architecture executes the pipeline normally while a parallel orchestrator perturbs each layer's intermediate state and measures the resulting shift in the final output via cosine divergence, producing an experimentally grounded causal weight per layer. Across 7 tasks and 42 perturbation experiments, the framework detects systematic confabulation — self-explanations that are causally disconnected from the layer's actual influence on the output — most notably in the final evaluator layer, which explains its output with full confidence despite having near-zero measured causal contribution.
Hierarchical Explanation Graph (HEG)
Causal explainability framework for hierarchical multi-agent LLM pipelines, detecting confabulated self-explanations via controlled perturbation experiments.
Ciffeer
AI agent SaaS platform designed and built solo, under own entity (Corarea). Multi-layer planning architecture — Orchestrator → Planner → Mid-Level Planner → Analyser → Executor → Evaluator. 100+ tools across 15 categories, DeepSeek as LLM backbone, semantic memory via pgvector, deployed on Google Cloud Run.
COLFI Back-Office Platform & LLM Monitoring Agent
React + FastAPI back-office platform for corporate action processing, OCR-assisted document extraction, ISO 20022 payload generation, and treasury funding allocation optimization. Contribution centered on an LLM monitoring agent for detecting and narrating drift in financial models — PPCA-based drift detection, a deterministic cash placement optimizer, and a governed recommendation schema.
ColFi Collateral Finance Management DApp
Decentralized asset and collateral management platform for financial institutions. Hyperledger Fabric chaincode for asset state transitions, collateral locking/unlocking, and immutable audit trails, with a Next.js frontend and a Fabric network hosted directly on Amazon EC2, backend via AWS Amplify.
Insight-Ray
AI-powered chest X-ray analysis system detecting and classifying 14 thoracic abnormality classes using a YOLOv8 model fine-tuned on the VinBigData Chest X-ray Abnormalities Detection dataset (18,000+ images). Combines detection with natural-language explanations via Google's Gemini API. Research/educational tool, not for clinical use.
Dalil AI
Offline-first desktop research platform unifying scikit-learn, PyTorch, TensorFlow, MediaPipe, and Stable-Baselines3/Gymnasium behind a single GUI. 7-step scikit-learn wizard, PyTorch transfer-learning lab, TFLite export for edge deployment, real-time MediaPipe vision, and an RL Studio for training PPO/DQN/SAC agents.
Breiman CART
Pure Python, from-scratch reimplementation of the original CART algorithm as specified by Breiman, Friedman, Olshen, and Stone (1984) — cost-complexity pruning, native categorical feature handling, and full tree inspection. Published as an installable package.
Vehicle Trajectory Prediction
Trajectory prediction model combining point-cloud feature extraction (PointNet) with a BiLSTM-Attention sequence model, benchmarked on the nuScenes dataset. Builds on prior CARLA-simulation-based autonomous vehicle work. Manuscript in preparation.
DApp Developer Intern
Built a decentralized asset and collateral management platform on Hyperledger Fabric. Chaincode for asset state transitions, collateral locking/unlocking, and immutable audit trails, with a Next.js frontend and a Fabric network hosted directly on Amazon EC2, backend via AWS Amplify (API Gateway, Lambda, DynamoDB).
LLM Monitoring Agent
Contributed an LLM-based financial model monitoring agent, built on top of the platform's existing multi-agent orchestration system, for detecting and narrating drift in financial models — including PPCA-based drift detection, a deterministic cash placement optimizer, and a governed recommendation schema.
Winner with Siana for developing an autonomous robotic system for high-speed train bogie inspection.
Awarded at the Fès-Meknès Economic Forum Hackathon for developing Ciffeer, an AI multi-agent platform.
Based in Meknès, Morocco.