Aicha Oukdour

Designed and developed an end-to-end AI coding assistant leveraging LLMs, RAG architectures (naive, advanced), and autonomous reasoning workflows. Fine-tuned LLMs/SLMs using LoRA, QLoRA, and Unsloth, improving model efficiency and reducing computational cost. Applied RLHF and DPO techniques to align models with user intent and enhance response quality. Built and optimized retrieval pipelines using advanced embedding models (BERT, Sentence Transformers) and vector databases (FAISS, Chroma) with hybrid search strategies. Implemented data preprocessing pipelines including data cleaning, chunking, splitting, and feature engineering for structured and unstructured datasets. Developed ETL pipelines and AI-driven data crawling (Crawl AI) to ingest and process large-scale datasets. Deployed scalable MLOps pipelines (MLflow, CI/CD) for model versioning, monitoring, and performance tracking (latency, errors). Optimized LLM inference performance (latency, cost, throughput) and ensured system scalability in production environments. Integrated secure, API-based architectures for seamless deployment of AI services.

Mise en place d'un système de contrôle qualité dans un environnement Big Data. Technologies : Hadoop, Cloudera, Kafka/NiFi, Spark, Python, SQL, Apache Superset.

Analysis of SaaS Architectures and Cloud Infrastructures. Integration with the team to study SaaS deployment models.