Recap and Feedback
In this part, we moved from plain prompt-based applications to systems that can use external knowledge more deliberately.
The main ideas were:
- embeddings make similarity-based retrieval possible,
- RAG lets an application ground answers in local documents,
- context strategy is a design choice rather than a fixed recipe,
- and evaluation turns vague impressions into explicit engineering feedback.
The tutorial also showed that a useful RAG application includes document preparation, retrieval logic, structured outputs, and an evaluation workflow that can be rerun after changes. The framework-variant chapter then showed that LangChainJS can shorten some of the implementation while leaving the main engineering decisions in place.
These ideas lead directly to the next set of questions. Once an application can use tools, retrieved context, and local documents, new risks appear:
- What if retrieved content contains malicious instructions?
- What if the system exposes data too broadly?
- What if the application automates decisions that should remain under human control?
The final part of the course focuses on those questions.
Next, please take a moment to reflect on your work in this part and provide feedback. Your input helps us improve the course materials.