I’m a Master’s student in Data Science (AI & LLM engineering) based in Casablanca.
I design and deploy reliable AI systems—from RAG over complex documents to agentic pipelines and cloud-native data products.

Currently open to: End-of-studies internship (Feb 2026).

• RAG & Knowledge Systems: embeddings, semantic chunking, reranking, validation, traceability
• NLP for Real-World Docs: NER pipelines, structuring + indexing, entity-centric retrieval
• Agentic Workflows: multi-agent orchestration, stateful context, routing + fallbacks
• Production Engineering: Docker, CI/CD, FastAPI, GCP deployment patterns

• 98% retrieval precision on legal-document RAG through semantic chunking + tuning
• 4× memory compression via scalar quantization (cost/perf optimized)
• 75% faster deployments with containerization + automated CI/CD on GCP

AI/ML Engineer Intern — Morosoft Solutions (Casablanca) (Jul 2025 → Sep 2025)

• Built an automated legal-document analysis RAG system to replace manual search
• Designed NER → knowledge base pipeline for structured, queryable legal entities
• Added LLM routing + fallback to reduce API cost and improve availability
• Containerized and deployed on GCP with CI/CD

A high-reliability assistant designed for the CAN2025 experience: instant answers, planning help, and grounded guidance.

• RAG + agentic workflow to handle match/venue info, logistics, FAQs, and planning requests with stateful context
• Multimodal: understands images/screenshots (posters, schedules, tickets, maps) and uses them as context
• Reliability by design: source-grounded answers, validation checks, and fallbacks when confidence is low
• Stack: Python · FastAPI · LLM routing · embeddings/vector search · GCP

A chatbot for a banking app that answers questions about bank guidelines, procedures, and applicable rules, while enabling permissioned analytics.

• Policy & knowledge assistant: Q&A over internal guidelines/laws with traceable, context-aware retrieval
• Consent-based analytics: provides insights from user data only with explicit permission (privacy-first by default)
• Security controls: least-privilege tool access, PII minimization/redaction, audit-friendly logging, secure data handling
• Multimodal: works with documents/images (e.g., statements, screenshots) to extract context before responding
• Stack: Python · RAG · NLP · FastAPI · secure orchestration · cloud deployment patterns

• Multi-agent system that builds & maintains knowledge graphs from unstructured sources
• Improves traceability and factual consistency beyond pure vector search using stateful reasoning + adaptive recovery

• Scraping + unification pipeline with Pydantic schemas for clean, consistent profile objects
• Hybrid matching (semantic similarity + keyword/rule signals) for accurate cross-platform identity linking ss components feeding analytics-ready storage and dashboards.

Languages: Python · SQL · Bash · Java
ML/NLP: PyTorch · Transformers · Hugging Face · Scikit-learn · NER · Quantization
Data: BigQuery · PySpark · Hadoop · Pandas · Data validation (Pydantic) · Web scraping
Vector/DB: Neo4j · Qdrant · FAISS · PostgreSQL · MongoDB · MySQL
Delivery: FastAPI · Docker · CI/CD · GCP · Gradio/Streamlit

• PyTorch for Deep Learning (DeepLearning.AI)
• CS50 AI with Python (HarvardX)
• Elements of AI for Business (MinnaLearn)
• English (C2) · French (B2)

• Email: aymen.echchalim@gmail.com
• LinkedIn: linkedin.com/in/echchalim
• Location: Casablanca, Morocco