CAIC AI Tools & Technologies 2 — Questions and Answers
Question 1: Which vector database is most commonly used to store and retrieve embeddings in Retrieval-Augmented Generation (RAG) pipelines?
- MySQL
- Pinecone (Correct answer)
- Redis Streams
- Apache Kafka
Correct answer: Pinecone
Pinecone is a purpose-built vector database widely used in RAG pipelines to store and perform similarity searches on embeddings.
Question 2: What is the primary role of LangChain in AI application development?
- Training large language models from scratch
- Orchestrating LLM calls, memory, and tool integrations into chains (Correct answer)
- Providing GPU infrastructure for model inference
- Monitoring model drift in production
Correct answer: Orchestrating LLM calls, memory, and tool integrations into chains
LangChain is a framework that orchestrates LLM calls, memory management, and external tool integrations into composable chains.
Question 3: A client wants to deploy an open-source LLM on their own servers to avoid data leaving their network. Which tool best supports this use case?
- OpenAI API
- Ollama (Correct answer)
- Google Vertex AI
- Amazon Bedrock
Correct answer: Ollama
Ollama allows users to run open-source LLMs locally or on private servers, keeping data fully on-premises.
Question 4: Which technique in prompt engineering explicitly instructs a model to reason step-by-step before producing its final answer?
- Few-shot prompting
- Zero-shot prompting
- Chain-of-thought prompting (Correct answer)
- Instruction tuning
Correct answer: Chain-of-thought prompting
Chain-of-thought prompting encourages the model to show intermediate reasoning steps, improving accuracy on complex tasks.
Question 5: What does the 'temperature' parameter control in LLM inference?
- The hardware power usage during generation
- The randomness or creativity of model outputs (Correct answer)
- The maximum number of tokens generated
- The model's training learning rate
Correct answer: The randomness or creativity of model outputs
Temperature scales the probability distribution over next tokens; higher values produce more random, creative outputs while lower values produce more deterministic ones.
Question 6: Which AI observability tool is specifically designed to trace, evaluate, and debug LLM applications end-to-end?
- Grafana Loki
- Prometheus
- LangSmith (Correct answer)
- Datadog APM
Correct answer: LangSmith
LangSmith, built by the LangChain team, provides tracing, evaluation, and debugging specifically for LLM-powered applications.
Question 7: When using OpenAI's API, which parameter limits the total number of tokens in the model's response?
- top_p
- max_tokens (Correct answer)
- frequency_penalty
- presence_penalty
Correct answer: max_tokens
The max_tokens parameter caps the number of tokens the model can generate in a single response.
CAIC AI Tools & Technologies 2
Which vector database is most commonly used to store and retrieve embeddings in Retrieval-Augmented Generation (RAG) pipelines?