Interview Questions and Final Revision

Agentic AI refers to AI systems that can autonomously plan, take actions, use tools and achieve goals with minimal human intervention. It goes beyond simple Q&A and can reason, decide and act in real world environments.

• RAG is mainly for knowledge retrieval and better response generation.
• Agentic AI can plan, use multiple tools, take actions and complete multi-step tasks.
• RAG can be a tool inside an agent.

1. 1LLM (reasoning and decision making)
2. 2Memory (short-term and long-term)
3. 3Tools (APIs, DB, search, etc.)
4. 4Planning (task decomposition)
5. 5Execution (action using tools)
6. 6Feedback loop (review and improve)

1. 1Understand the task / goal
2. 2Plan the steps
3. 3Select and use appropriate tools
4. 4Execute the actions
5. 5Observe the results
6. 6Re-plan if needed
7. 7Give the final answer

• Web search (Google, SerpAPI)
• APIs (weather, flight, payment, etc.)
• Databases (SQL, MongoDB)
• File operations (read / write / upload)
• Code execution (Python REPL)
• Custom business tools (internal APIs)

• Define clear tool descriptions and schema.
• Use function calling or tool calling.
• Validate inputs and handle errors.
• Implement retries and fallback tools.
• Log tool usage for observability.

• Task completion rate
• Correctness of output
• Tool usage accuracy
• Latency and cost
• Human feedback / user satisfaction

• Hallucination and incorrect actions
• Tool selection and planning errors
• Handling long and complex tasks
• Security and safety issues
• Cost and latency
• Evaluation is difficult

• LangChain (tools, agents, memory)
• LlamaIndex (data and RAG + agents)
• AutoGen (multi-agent systems)
• CrewAI (role based agents)
• OpenAI Agents (tool calling)
• LangGraph (advanced agent workflows)