Gen AI & Agentic AI Course in Pune

•  What is NLP? Real-World Applications
•  What is Generative AI? Use Cases and Trends
•  Evolution of NLP: From Rule-Based to Deep Learning Approaches
• Overview of Generative AI Models

1. Text Preprocessing:

• Tokenization (Word, Sentence, Subword Tokenization)
• Lemmatization and Stemming
• Stop Words Removal

2. Part-of-Speech (POS) Tagging

3. Named Entity Recognition (NER)

1. Traditional Representations

• Bag of Words (BoW)
• TF-IDF

2. Dense Word Representations:

•  Recurrent Neural Networks (RNNs)
•  Long Short-Term Memory (LSTM) and Bidirectional LSTMs
• Gated Recurrent Units (GRU)
• Applications of RNNs, LSTMs, and GRUs in NLP

1. Introduction to Transformers:

•   Why Transformers were developed.
•   Limitations of RNNs and LSTMs.

2. Attention Mechanisms:

• Self-Attention: Key, Query, and Value Vectors, Scaled Dot-Product Attention.
• Multi-Head Attention: Capturing diverse relationships within sequences.
• Advanced Attention Mechanisms:
• Multi-Query Attention (MQA)
• Grouped-Query Attention (GQA)
• Sliding Window Attention (Longformer)
• Flash Attention (Efficient GPU-Optimized Attention)

1. Transformer Encoder-Decoder Architecture

2. Positional Encoding: Sinusoidal vs. Learnable

3. Encoder Components: Self-Attention, Feedforward Network, Layer Normalization

4. Decoder Components: Masked Self-Attention, Cross-Attention

Variants of Transformers

• Encoder-Only Models (BERT, RoBERTa)
•  Decoder-Only Models (GPT, GPT-4)
•  Encoder-Decoder Models (T5, Seq2Seq)

5. Applications: Machine Translation, Summarization, Text Generation, Knowledge Retrieval

1. What Makes a Model Generative?

2. Autoregressive (GPT) vs. Autoencoding (BERT)

3. Large Language Models (GPT-4, PaLM, Claude, LLaMA)

4. Popular Pretrained Models:

•    BERT, RoBERTa, GPT, T5, XLNet

5. Transfer Learning and its Importance

6. Tools: Hugging Face Transformers, spaCy, NLTK

1. Full Fine-Tuning vs. Partial Fine-Tuning

2. Parameter-Efficient Fine-Tuning:

•  LoRA, QLoRA, Adapters, Prefix Tuning, Prompt Tuning

3. Domain-Specific Fine-Tuning (Healthcare, Finance, Legal AI)

4. Few-Shot and Zero-Shot Fine-Tuning

5. Model Evaluation & Explainability:

•  Perplexity (Measuring model uncertainty in predictions)
•  BLEU, ROUGE, METEOR (Text generation quality)
•  BERTScore (Semantic similarity-based evaluation)
•  Exact Match (EM), F1-Score (QA and retrieval-based evaluations)
•  Human Evaluation & GPT-Assisted Evaluation

6. Challenges in Fine-Tuning:

•  Overfitting in Small Datasets
•  Catastrophic Forgetting in Domain-Specific Training
• Handling Large Models with Limited Compute

1. What is Prompt Engineering?

2. Importance of Prompts in LLMs

3. Designing Effective Prompts

4. Few-Shot, Zero-Shot, and Chain-of-Thought Prompting

5. Optimizing Prompts:

• Self-Consistency & Iterative Refinement

 6. Dynamic Prompting with LangChain

1. What is RAG?

2. Role of RAG in Generative AI Workflows

3. Building RAG Pipelines:

• Query Embedding Extraction
• Hybrid Search (BM25 + Dense Retrieval)

4. Chunking Techniques:

• Fixed-Length, Semantic, Recursive

5. Vector Databases:

•  ChromaDB, Milvus, Pinecone, FAISS, CosmosDB

6. Re-Rankers: Improving Retrieval Results

7. Challenges in RAG Implementation:

•       Indexing Latency
•       Hallucination Risk
•       Computational Cost
• Privacy & Security Considerations

1. Multi-Modal Models (DALL-E, CLIP, Flamingo)

2. Reinforcement Learning with Human Feedback (RLHF)

3. Adaptive Fine-Tuning & Continuous Learning

4. Generative Adversarial Networks (GANs)

1. What is Agentic AI?

2. Difference Between Task-Oriented AI and Autonomous AI

3. Core Components of Agentic AI:

•  Decision-Making Capabilities
•  Feedback Loops & Self-Learning
•  Dynamic Adaptation to Environments

4. Advanced Agentic AI Techniques:

•  ReAct Framework
•  Multi-Agent AI Systems
•  CrewAI for Task Delegation
•  LangGraph: Graph-Based Workflow Design

5. Applications of Agentic AI

• Task Automation with AI Agents
• Multi-Agent Systems in Gaming, Healthcare, and Logistics
• Collaborative Systems for Complex Workflows

1. Amazon S3: Storage for datasets, model checkpoints

2. Amazon SageMaker: Train, fine-tune, and deploy LLMs

3. AWS Bedrock: Access prebuilt foundation models

4. AWS Lambda: Serverless inference for LLMs

5. AWS Textract: Extract text, forms, and tables from documents