No description available

No description available

project on llm and rag application

Developed a DataOps platform to ingest and preprocess scalable, multimodal data, using a layer-based architecture stored in MinIO over Docker. An LLM/ViT transformer model (CLIP) was integrated to extract embeddings for both text and images in the same vectorial space. Using Hugging Face, this model was efficiently fine-tuned with PEFT (LoRA/QLoRA)

End-to-end DataOps platform for monitoring pipeline failures, schema changes, SLA breaches, and downstream impact using Kafka, Airflow, PostgreSQL, dbt, Superset, and RAG.