IVQ 201-250

Section 21: Scaling GenAI Systems (10 Questions)

1. How do you horizontally scale a GenAI service?

2. What are GPU memory optimization strategies for LLM inference?

3. How would you handle rate limits for large-scale OpenAI API usage?

4. What’s the difference between multi-GPU vs. multi-node LLM serving?

5. How can you use model sharding in production environments?

6. What is speculative decoding and how does it improve throughput?

7. What are the main bottlenecks in high-load GenAI applications?

8. How do you manage logs and observability in a GenAI backend?

9. How do you build a cost-monitoring dashboard for GenAI endpoints?

10. What is KV cache reuse and how does it optimize performance?

---

Section 22: Agent Architectures & Tool Use (10 Questions)

1. What are the main types of agents in LangChain or AutoGen?

2. How do agents handle tool selection dynamically?

3. What’s the difference between planner-executor and reactive agents?

4. How would you design a memory-aware GenAI assistant using tools?

5. How can agents collaborate or hand off tasks to each other?

6. How do you sandbox tool-executing agents for safety?

7. How do you create an autonomous agent for report generation?

8. What role does the scratchpad play in reasoning agents?

9. How do agents balance exploration vs. exploitation in decision-making?

10. How can GenAI agents perform complex multi-step workflows?

---

Section 23: Personalization & Adaptive GenAI (10 Questions)

1. How do you personalize a GenAI chatbot for individual users?

2. What’s the role of user embeddings in personalizing responses?

3. How do you maintain personalization across sessions securely?

4. What are privacy challenges in personalized GenAI apps?

5. How do you use retrieval to provide contextually aware responses?

6. How do you fine-tune an LLM on user feedback?

7. What is reinforcement learning with user signals (RLUS)?

8. How can GenAI recommend content adaptively in real time?

9. What’s the tradeoff between personalization and generalization?

10. How would you build a personalized study tutor using GenAI?

---

Section 24: Alignment, Bias & Control (10 Questions)

1. What are the principles behind Constitutional AI?

2. How do you ensure alignment of LLMs with company values?

3. How can LLMs be aligned post-deployment?

4. What is the difference between supervised fine-tuning and RLHF?

5. How do you detect harmful or biased outputs in real-time?

6. What are adversarial prompts, and how do you defend against them?

7. How can GenAI support explainability in its outputs?

8. What’s the role of safety layers and filters like OpenAI’s Moderation API?

9. How do you manage trade-offs between safety and creativity in generation?

10. How do companies like Anthropic approach LLM alignment?

---

Section 25: GenAI in Low-Resource & Edge Environments (10 Questions)

1. How do you compress models for edge deployment?

2. Compare TinyLLaMA, DistilGPT, and other small LLMs.

3. How do you handle low-bandwidth GenAI applications?

4. What are on-device privacy benefits for GenAI assistants?

5. How do you optimize inference for ARM architectures?

6. What’s the best model quantization method for small devices?

7. How do you deploy a GenAI chatbot on a Raspberry Pi?

8. What are trade-offs when using 4-bit vs. 8-bit quantized models?

9. What are low-resource strategies for building domain-specific GenAI apps?

10. How do you use LoRA for quick personalization in resource-constrained environments?

---