To Nha Notes | April 28, 2025, 2:59 p.m.
In this project, you built a full Retrieval Augmented Generation (RAG) system by combining a Kaggle dataset, a Hugging Face language model (like TinyLlama), and ChromaDB as a vector database. You learned how to store embeddings, query relevant documents based on vector distances, and answer user questions with high relevance.
To push your skills further, you are encouraged to:
Try different datasets and adapt your questions to the new data.
Experiment with alternative Hugging Face models, understanding trade-offs like performance vs. memory usage.
Replace ChromaDB with other vector databases like Pinecone or Weaviate to deepen your architectural knowledge.
Each of these exercises strengthens your ability to customize and optimize AI pipelines. While swapping vector databases is more complex, it offers valuable real-world experience working with modern AI tech stacks.
You’ve also gained foundational knowledge on Hugging Face libraries, explored what RAG systems are, and practiced setting up end-to-end pipelines. These skills are essential for building more advanced LLM applications in the future.
In the next step, you’ll dive into LangChain to further enrich your RAG systems and explore intelligent agents powered by LLMs.
Pere Martra, Large Language Models Projects: Apply and Implement Strategies for Large Language Models (Apress)