Font size
WorksheetsAgentic AI Internship
Total questions: 63
Worksheet time: 1hrs 3mins
In LangChain's schema, what does the 'Model' component represent?
A specific dataset used for training.
The core reasoning engine, like GPT-4 or Google's Gemini.
The final output structure of the application.
A pre-written prompt.
How does an 'Output Parser' enhance an LLM's response?
It checks the LLM's response for factual accuracy.
It converts the LLM's raw text output into a structured format like a list or JSON.
It shortens the LLM's response to save on tokens.
It adds more creativity to the LLM's output.
What is the most fundamental definition of a 'Chain' in LangChain?
A sequence of operations where the output of one step becomes the input for the next.
A security protocol for protecting API keys.
A method for connecting multiple LLMs together.
The graphical user interface for a LangChain application.
Why is it a best practice to store your API key in a .env file?
It makes the key easier to remember.
To keep the key safe and out of your source code, preventing accidental exposure.
The langchain library requires it to function.
It allows the API key to be changed dynamically without restarting the program.
What does setting verbose=True in an LLMChain do?
It makes the final output from the LLM longer and more detailed.
It prints the internal steps of the chain, like the fully formatted prompt, which is useful for debugging.
It automatically adds comments to your Python code.
It enables a more creative (higher temperature) response from the LLM.
A simple LLMChain follows which visual flow?
[User Input] -> [LLM] -> [PromptTemplate] -> [Output]
[User Input] -> [PromptTemplate] -> [LLM] -> [Output]
[PromptTemplate] -> [User Input] -> [LLM] -> [Output]
[LLM] -> [PromptTemplate] -> [User Input] -> [Output]
What is the single most important reason that chatbots need a 'memory' system?
To make their responses faster.
Because LLMs are stateless and have no memory of past interactions by default.
To save the conversation to a text file automatically.
To reduce the cost of API calls.
Which of the following is a potential drawback of ConversationBufferMemory?
It is very difficult to implement.
It can lose important context from the beginning of a long conversation.
It stores a summary, which can be inaccurate.
Its memory can eventually become too large and exceed the LLM's context window.
According to the lesson, what is the modern, industry-standard way to manage state in LangChain for complex applications?
Using a simple Python list to store messages.
Using ConversationBufferWindowMemory.
Using LangGraph to explicitly define and control the application's state.
Writing the conversation history to a CSV file.
In the LangGraph example, what is the purpose of the AgentState class?
To define the structure of the data that is passed between the nodes of the graph.
To store the OpenAI API key securely.
To choose which LLM to use.
To execute the final step of the graph.
What is a 'checkpointer' (like MemorySaver) used for in LangGraph?
To verify the factual accuracy of the LLM's response.
To control the flow and logic between nodes.
To persist (save) the state of the graph, allowing conversations to be resumed.
To time how long each node takes to execute.
In the LangGraph code, what does the line workflow.add_edge('agent', END) signify?
It signifies that after the 'agent' node runs, the graph's execution should terminate.
It connects the 'agent' node to a tool.
It indicates an error in the 'agent' node.
It marks the beginning of the graph's execution.
Why is a unique thread_id important when invoking a stateful LangGraph application?
It tells the LLM which user is talking.
It allows the memory system to track and save multiple conversations independently.
It is a security feature to prevent unauthorized access.
It helps in debugging by labeling the output.
How does LangGraph offer a more scalable approach than simple memory objects?
It uses less computer memory.
It is designed for complex agents that need to manage memory, tools, and decision-making in an organized way.
It only works with the most powerful LLMs.
It automatically summarizes the conversation history.
What is the fundamental principle of how chatbot memory works?
The LLM's internal state is modified with each turn.
The entire conversation history (or a part of it) is manually included with every new user query.
The chatbot uses a separate, smaller LLM just for remembering things.
The LLM's weights are fine-tuned in real-time during the conversation.
The LangGraph pattern shifts the developer's focus from just 'memory' to the broader concept of...?
Application 'State'.
API 'Calls'.
User 'Interface'.
Prompt 'Engineering'.
What are the two main limitations of LLMs that 'Tools' help overcome?
Slow response times and high cost.
Inability to understand foreign languages and complex grammar.
Knowledge cut-off dates and unreliability in performing precise calculations.
Limited creativity and difficulty in writing long-form text.
What is the core function of a LangChain 'Agent'?
To simply pass a user's prompt to an LLM.
To use an LLM to make decisions about whether to use a tool, and which tool to use.
To compress the conversation history into a summary.
To translate the user's input into different languages.
The 'ReAct' framework allows an agent to...
...react emotionally to a user's input.
...generate a final answer in a single step.
...enter a loop of internal Reasoning and Acting (using tools) until it finds a solution.
...rewrite the user's prompt to be more efficient.
In the ReAct cycle, what is an 'Observation'?
The agent's internal thought about which tool to use.
The final answer given to the user.
The output or result the agent receives after using a tool.
The initial prompt from the user.
Why would an agent choose the PythonREPLTool to solve 'What is (35.2 * 14.9) / 2.5?'
Because the question involves Python programming.
Because LLMs are unreliable at mathematical calculations, and the tool can execute the math precisely.
Because the Wikipedia tool cannot handle numbers.
Because it's the only tool available.
What is the role of the AgentExecutor?
It is the LLM that powers the agent's reasoning.
It is the prompt template that instructs the agent.
It is the runtime environment that executes the agent's reasoning loop, calls tools, and returns the final response.
It is the tool that the agent uses.
To use a tool like TavilySearchResults, what must the user typically do first?
Install the langgraph library.
Get a specific API key for that tool and set it as an environment variable.
Increase the LLM's temperature setting.
Define a custom output parser.
What is the main advantage of giving an agent access to a tool like Wikipedia?
It allows the agent to bypass its knowledge cut-off date and access current information.
It makes the agent's responses much shorter.
It guarantees every answer will be 100% correct.
It reduces the number of tokens used in the prompt.
Pulling a prompt from the LangChain Hub (e.g., hub.pull('hwchase17/react')) is beneficial because:
It's the only way to create a ReAct agent.
It allows you to use a battle-tested, pre-built prompt designed specifically for a complex task.
It ensures your agent will use the latest OpenAI model.
It provides free access to the tools mentioned in the prompt.
Giving an agent tools fundamentally transforms it from a 'language model' into a...?
Database.
User interface.
Problem-solving engine.
Chat history logger.
1. What does MCP stand for?
Model Command Protocol
Model Context Protocol
Machine Control Protocol
Multi-Client Platform
2. What is MCP often compared to in terms of its role in AI applications?
A smartphone
A programming language
A USB-C connector
An AI assistant
3. What problem does MCP solve in AI integration?
Building new AI models
Complex and redundant model-tool connections
Training datasets faster
Increasing GPU speed
4. In the MCP architecture, what is the main role of the Host?
Executes tool actions
Provides UI design
Manages conversation and decides which tools to call
Translates code
5. Which of the following is NOT one of the MCP key benefits?
Dynamic Tool Discovery
Multi-step Orchestration
Interoperability
Auto-model training
6. What is the role of the MCP Client in the architecture?
Hosts the AI model
Sends and receives MCP requests/responses
Decides which tool to call
Stores user data
7. What are MCP Servers responsible for?
Exposing tools, resources, and prompts in a standard format
Writing AI code
Hosting LLMs
Translating between languages
8. How does MCP simplify the M×N integration problem?
By reducing memory usage
By connecting each model and tool only once via a shared protocol
By using faster processors
By translating AI languages
9. Which transport mechanisms does MCP support?
USB and Bluetooth
HTTP and FTP
XML and JSON-RPC
Stdio and HTTP + SSE
10. What are the three main types of "Server Primitives" in MCP?
Resources, Tools, Prompts
Tools, APIs, Services
Documents, Scripts, Models
Requests, Commands, Outputs
How does specificity improve AI outputs?
Specificity reduces the amount of data processed by AI.
Specificity improves AI outputs by enhancing relevance and accuracy through tailored responses.
Specificity makes AI outputs more general and less focused.
Specificity has no impact on the quality of AI responses.
Provide an example of a specific prompt.
Write a short story about a dragon who learns to fly.
Explain how to train a dragon.
Describe a dragon's favorite food.
List the colors of different dragons.
What does role assignment in prompt engineering entail?
Creating a list of keywords for the AI to use.
Limiting the AI's responses to yes or no answers.
Defining specific roles or personas for the AI to adopt in generating responses.
Assigning random tasks to the AI without context.
Give an example of a role assignment prompt.
As a software developer, write a user manual for the application.
As a project manager, outline the key milestones for the upcoming project.
As a data analyst, summarize the quarterly financial report.
As a graphic designer, create a marketing strategy for the product.
What is the purpose of using constraints in prompts?
To increase the length of the response.
To guide and limit the response to be more relevant and focused.
To enhance creativity in responses.
To allow for more diverse interpretations.
How can you summarize information effectively in prompts?
Avoid summarizing and provide full paragraphs.
Identify key points and use concise language.
Use complex jargon and lengthy explanations.
Focus on irrelevant details and examples.
What does ELI5 stand for in prompt engineering?
Explain Like I'm 10
Every Little Idea 5
Explain Like I'm 5
Engage Learning in 5
Provide an example of an ELI5 prompt.
Describe a cat as if I'm 5.
Tell me how to cook spaghetti like I'm 5.
Explain the internet like I'm 5.
Explain gravity to a toddler.
What is the chain-of-thought technique?
A strategy that emphasizes group decision-making over individual reasoning.
A technique that focuses on memorizing facts without analysis.
A method of reasoning that relies solely on intuition.
The chain-of-thought technique is a method of reasoning that involves sequentially breaking down a problem into smaller steps.
How does the chain-of-thought technique benefit AI responses?
It improves reasoning and accuracy in AI responses.
It limits the scope of AI understanding.
It reduces the amount of data processed by AI.
It decreases the complexity of AI algorithms.
What are negative constraints in prompt engineering?
Negative constraints are optional guidelines for prompt flexibility.
Negative constraints are suggestions for improving prompt clarity.
Negative constraints are guidelines that specify what should not be included in the output.
Negative constraints are rules that enhance the output quality.
How can negative constraints improve the quality of AI outputs?
Negative constraints improve AI output quality by preventing undesirable content generation.
Negative constraints are only useful for data storage.
Negative constraints have no impact on AI performance.
Negative constraints limit creativity in AI outputs.
What is a pro tip for enhancing clarity in prompts?
Use vague language and avoid details.
Ask multiple questions at once.
Provide irrelevant information.
Be specific and provide context.
Asher is working on a project that involves creating a chatbot. He wants to improve the chatbot's responses by using information from various online sources. What is this technique called?
A method for training language models using only internal data.
A way to generate text without any external information.
A technique that enhances language models by incorporating external knowledge sources.
What are the two key principles that RAG combines?
Information Retrieval and Language Generation
Data Mining and Text Analysis
Machine Learning and Data Science
A medical RAG chatbot provides correct answers but in a very generic style. The hospital wants the tone to be “professional and empathetic.” What’s the best approach?
Fine-tune the retriever on empathetic text
Add prompt engineering on top of RAG to shape response style
Replace RAG with supervised fine-tuning
Reduce retrieval k to get more specific documents
A major limitation of RAG compared to fine-tuning is:
RAG requires a static knowledge base
RAG cannot adapt to unseen queries
RAG depends on retrieval latency and quality of external documents
RAG cannot generate long-form answers
Chunking in RAG is used mainly to:
Reduce LLM latency
Prevent overfitting of retrievers
Break documents into manageable pieces for retrieval
Increase model interpretability
What is semantic similarity search?
Finding documents that match keywords exactly.
A method for generating random text.
Retrieving information/documents that are semantically closest to the user's query.
What is the main difference between semantic search and dense retrieval?
Semantic search focuses on understanding meaning, while dense retrieval focuses on mapping text into a vector space.
Dense retrieval is faster than semantic search.
Semantic search uses more data than dense retrieval.
Compared to end-to-end fine-tuning, RAG is more suitable when:
The task is closed-domain with limited vocabulary
The knowledge base is frequently updated
The model must be highly parameter-efficient
Training compute is unlimited
Your RAG pipeline often returns too much irrelevant information. Which change improves retrieval precision without losing recall?
Use random sampling for diversity
Increase k (top retrieved results)
Reduce chunk size and use better embedding model
Disable normalization of embeddings
A RAG-based summarizer for legal contracts produces overly long answers. What’s the best fix?
Reduce temperature
Increase k to retrieve more context
Use larger embeddings
Prompt the LLM with “Answer in under 100 words” after retrieval
Enter Your Name
Enter Enrolment No (Use Capital Alphabets Only)
Enter Your Branch From this(CSE,CSE-AIML,CSE-DS,CSIT,CY,IT)
Enter Your Acropolis Email ID( For Ex xxxxxxxxxx@acropolis.in)
