Generative AI Training in Perungudi

• Introduction to Artificial Intelligence, Machine Learning, and Generative AI
• Python programming for AI development
• Essential mathematics for neural networks
• Deep learning fundamentals and neural network concepts
• Evolution of Natural Language Processing (NLP) into Generative AI
• Core tools and frameworks: Python, NumPy, PyTorch, and TensorFlow

Beginner → Intermediate Build: Implement a foundational neural network from scratch using Python and NumPy.

• Introduction to Large Language Models and their applications
• Transformer architecture, attention mechanisms, encoders, and decoders
• Tokenization, embeddings, and text generation
• Understanding GPT, BERT, and LLaMA model architectures
• How LLMs process prompts and generate responses
• Practical implementation with Hugging Face Transformers and PyTorch

Intermediate Build: Load, tokenize, and fine-tune a pre-trained Hugging Face Transformer model.

• Fundamentals of prompt engineering
• Zero-shot, few-shot, and Chain-of-Thought prompting
• ReAct prompting and structured output generation
• Managing context windows, temperature, and top-p settings
• Integrating LLMs into applications using APIs
• Hands-on integration with OpenAI, Anthropic Claude, and Google Gemini APIs

Intermediate → Advanced Build: Develop a multi-turn conversational AI application using an LLM API.

• Introduction to LLM application development
• Building sequential chains and reusable workflows
• Managing conversational memory and chat history
• Loading, splitting, and processing documents
• Developing question-answering applications
• Creating interactive AI interfaces

Intermediate → Advanced Build: Develop an interactive, multi-document Question & Answer application using LangChain.

• Understanding embeddings and vector representations
• Measuring vector similarity using cosine similarity and dot product
• Implementing semantic search and information retrieval
• Vector indexing and efficient retrieval techniques
• Storing and retrieving document embeddings

Intermediate Build: Implement a high-performance semantic search engine using Pinecone or ChromaDB.

• Introduction to RAG architecture and workflows
• Building document-based question-answering systems
• Advanced RAG techniques and retrieval optimization
• Implementing hybrid search using keyword and semantic retrieval
• Document re-ranking and contextual retrieval
• Processing complex PDFs, tables, and enterprise documents
• Evaluating retrieval quality and reducing hallucinations

Advanced Build: Architect an advanced RAG pipeline capable of parsing complex enterprise PDFs and tables, utilizing hybrid search.

• Introduction to model fine-tuning and customization
• Full fine-tuning versus parameter-efficient fine-tuning (PEFT)
• Implementing LoRA and QLoRA techniques for memory efficiency
• Preparing, cleaning, and curating high-quality training datasets
• Introduction to Reinforcement Learning from Human Feedback (RLHF)
• Evaluating fine-tuned models for performance and safety

Advanced Build: Fine-tune an open-source LLM (like LLaMA-3 or Mistral) on a custom domain-specific dataset using QLoRA.

• Introduction to multimodal AI applications and use cases
• Image generation using diffusion models and GANs
• Integrating text, image, and audio inputs into unified pipelines
• Speech recognition and high-fidelity audio generation
• Building end-to-end AI applications that process images and voice

Advanced Build: Create a multimodal application that transcribes meeting audio using Whisper, generates an AI summary.

• Introduction to autonomous AI agents and agent architectures
• Tool calling and function calling mechanisms
• Connecting AI agents to web search, APIs, calculators, and SQL databases
• Building agents with advanced planning, reasoning, and memory
• Designing multi-agent workflows and task orchestration
• Developing automated, end-to-end business workflows

Expert Build: Architect a multi-agent workflow using CrewAI or LangGraph where specialized agents collaborate to research a topic.

• Building robust APIs for Generative AI applications
• Deploying models and applications to production environments
• Containerization using Docker for consistent deployments
• Cloud deployment fundamentals (AWS, Google Cloud)
• Optimizing inference speed, latency, and resource usage
• Implementing caching, logging, and performance monitoring
• Tracking and debugging LLM application workflows

Advanced Build: Containerize and deploy a production-ready Generative AI API using FastAPI and Docker.

• Responsible AI principles, fairness, and bias mitigation
• Identifying and preventing prompt injection and jailbreak attacks
• Protecting sensitive data and masking personally identifiable information (PII)
• Implementing rigorous input and output validation
• Configuring AI guardrails and content filtering mechanisms
• Understanding privacy, security, and enterprise compliance frameworks

Advanced Build: Implement a secure AI interface that actively detects prompt injections.

• Developing end-to-end Generative AI applications
• Building an enterprise RAG chatbot or intelligent knowledge assistant
• Creating an autonomous AI agent or multi-agent workflow
• Integrating LLMs, vector databases, APIs, and external tools seamlessly
• Testing, evaluating, and deploying the final application
• Building a professional GitHub portfolio and documenting projects
• Resume preparation, technical interview practice, and career guidance

Expert Build: Architect, deploy, and document a fully functional, enterprise-grade Generative AI application.