Overview

The course offers over 1200 carefully curated multiple-choice questions covering all key areas of Generative AI and LLMs. Each question is accompanied by detailed explanations, ensuring learners understand not only the right answers but also the reasoning behind them.You will explore transformer architecture, attention mechanisms, pretraining vs. fine-tuning, prompt engineering, RAG (Retrieval-Augmented Generation), zero-shot/few-shot learning, LLM evaluation metrics, deployment strategies, and more.Topics Covered1. Transformer ArchitectureSelf-Attention MechanismScaled dot-product attentionMulti-head attentionQuery, Key, Value operationsPositional EncodingSinusoidal vs learnedResidual Connections and Layer NormalizationFeedforward layersEncoder vs DecoderCausal vs Bidirectional attentionMasked attention2. Pretraining Objectives of LLMsCausal Language ModelingMasked Language ModelingSpan CorruptionNext Sentence PredictionPrefix Language ModelingInstruction-style pretraining3. LLM Fine-Tuning TechniquesFull Fine-tuningLoRAQLoRAAdaptersPrefix TuningPrompt TuningPEFTInstruction TuningFLAN, T0, Dolly, AlpacaSFT4. Prompt EngineeringPrompt Design PrinciplesClear instructionsContext-aware phrasingZero-shot, One-shot, Few-shot promptingChain of Thought promptingSelf-Consistency DecodingReAct promptingPrompt Injection and JailbreaksAutoPrompt, Soft Prompts (Prompt Tuning)5. LLM Evaluation Metrics and TechniquesAutomatic EvaluationBLEU, ROUGE, METEOR, BERTScore, MoverScoreEmbedding-Based EvaluationCosine similarity, dot product in embedding spaceLLM-as-a-JudgeHuman EvaluationTruthfulness, coherence, relevanceHallucination detectionToxicity/Bias detection6. Decoding StrategiesGreedy DecodingBeam SearchTop-k SamplingTop-p (Nucleus) SamplingTemperature-based SamplingRepetition PenaltyContrastive DecodingMixture DecodingEvaluation of Fluency vs Diversity7. Embedding Models and Vector SearchEmbedding Generation ModelsSentence-BERTe5, GTE, InstructorOpenAI text-embedding-adaSimilarity MetricsCosine similarity, dot productVector StoresFAISS, Chroma, Weaviate, PineconeSearch MethodsDense retrievalSparse retrieval (BM25)Hybrid search8. Retrieval-Augmented GenerationChunking strategiesFixed-size, sliding window, recursive, semantic chunkingRetriever architectureVector-based, dense retrieversPrompt templates for RAGFusion-in-Decoder, FiD-RAGMemory-efficient RAGEvaluation of RAG pipelinesLatency, F1, RecallK, hallucination rate9. LLM AgentsAgent FrameworksLangChain AgentsLangGraph (State Machine)ReAct (Reason + Act)Tool use in LLMsCalculator, Search, APIsGuardrails and Error Handling10. Serving and Inference OptimizationQuantization8-bit, 4-bitGGUF formatKV CacheUsed for fast autoregressive decodingFlashAttention, xFormersDeepSpeed Inference, vLLMServing FrameworksTGI, Triton, vLLM, llama.cpp, Hugging Face Inference Endpoints11. Common LLM Failure ModesHallucinationsToken limit truncationPrompt injectionOverfitting during fine-tuningPoor RAG retrievalContext window exhaustion12. LLMOps Using AWSAnd Much More!Special emphasis is placed on interview readiness - making sure you're well-prepared for roles at top tech companies working with or on LLMs. You'll also learn about ethical concerns, AI safety, and hallucination mitigation, all of which are becoming essential in modern AI applications.Whether you're a data science professional or a student aspiring to work in NLP or AI research, this course provides a structured, engaging, and interview-focused learning experience and ace your complex scenario-based interview.

Skills

Reviews