A vector database stores embeddings, which are numerical representations of data, so a RAG system can retrieve semantically relevant information for grounding the model's response.

Regularization techniques help models generalize better and avoid memorizing training data.

Tokenization converts text into smaller units (tokens) that AI models can process.

Parameter count affects model size, memory usage, and compute requirements, so a smaller model is usually more suitable for a mobile application with limited local resources.

In the requirements phase, AI can analyze stakeholder documentation to identify logical gaps or inconsistencies and help convert business needs into structured user stories.