Generative AI Training in Sholinganallur

• AI, ML, Deep Learning, and NLP landscape overview.
• Transformer architecture and attention mechanisms explained simply.
• Tokens, context windows, and model parameters.
• Proprietary (OpenAI, Anthropic) vs. Open-Source (Llama 3, Mistral) models.
• Setting up the AI development environment (Python, VS Code, Conda).
• Git version control and secure API key management (.env).

Beginner Build: A secure, offline chat interface using Ollama and a local open-source model.

• Advanced Python: OOP, decorators, and async programming for AI.
• Handling JSON, external APIs, and data serialization.
• Robust error handling, logging, and debugging for AI apps.
• Data manipulation with Pandas and NumPy for AI datasets.
• Introduction to AI orchestration frameworks (LangChain / LlamaIndex).
• Building reusable, modular code structures for AI projects.

Beginner Build: An AI-ready data pipeline script that fetches, cleans, and structures raw data from an external API.

• Zero-shot, few-shot, and role-based system prompting.
• Prompt templating, variable injection, and workflow design.
• Generating structured outputs (JSON, Pydantic schemas) from LLMs.
• Controlling tone, format, and context length effectively.
• Prompt testing, iteration, and reducing hallucination rates.
• Token counting and cost optimization strategies for prompts.

Beginner → Intermediate Build: An automated data extractor that converts messy, unstructured text into validated JSON.

• Working with modern LLM APIs and streaming responses.
• Managing conversation history and short-term memory.
• Fundamentals of function calling and tool definitions.
• Connecting models to external APIs (e.g., Weather, Search, CRM).
• Building multi-step, tool-using AI workflows.
• Error handling and fallback mechanisms for failed API calls.

Intermediate Build: An AI task assistant that checks real-time data via API and executes multi-step user requests.

• Embeddings, semantic similarity, and vector search mechanics.
• Document loading, cleaning, chunking, and metadata strategies.
• Vector databases deep dive (Pinecone, Milvus, Chroma).
• Building a complete Retrieval-Augmented Generation (RAG) pipeline.
• Advanced RAG: Hybrid retrieval, query rewriting, and reranking.
• Evaluating RAG systems for relevance, faithfulness, and context precision.

Intermediate → Advanced Build: An Enterprise RAG Knowledge Helper that accurately answers questions based on uploaded company PDFs.

• Introduction to Vision-Language Models (VLMs).
• Processing and extracting text from PDFs, images, and scanned docs (OCR).
• Multimodal embeddings and cross-modal retrieval.
• Structuring data extracted from complex tables and charts.
• Combining vision models with RAG for smarter document search.
• Real-world use cases in legal, healthcare, and finance document intelligence.

Advanced Build: A multimodal document intelligence system that reads invoice images or PDFs and extracts structured billing data.

• Decision framework: Prompting vs. RAG vs. Fine-Tuning.
• Preparing, cleaning, and formatting high-quality fine-tuning datasets.
• Parameter-Efficient Fine-Tuning (PEFT): LoRA and QLoRA concepts.
• Hands-on fine-tuning of open-source models (e.g., Llama 3).
• Model evaluation metrics (BLEU, ROUGE, and LLM-as-a-Judge).
• Model quantization (4-bit/8-bit) for efficient local deployment.

Advanced Build: A domain-specific AI assistant by fine-tuning a small open-source model on a custom dataset.

• Agent architectures: ReAct (Reason + Act) and Plan-and-Execute.
• Short-term and long-term memory management for agents.
• Multi-agent collaboration and role assignment.
• Tool discovery, decision-making loops, and self-correction.
• Frameworks for agentic AI (LangGraph, AutoGen, CrewAI).
• Debugging and monitoring autonomous agent behavior.

Advanced Build: An autonomous AI research agent that plans queries, searches the web, summarizes findings, and saves reports.

• Introduction to Model Context Protocol (MCP) architecture.
• Building MCP servers and clients from scratch using Python.
• MCP core primitives: Tools, Resources, and Prompts.
• Secure authentication and controlled action execution.
• Agent-to-Agent (A2A) communication frameworks.
• Integrating MCP with existing enterprise APIs and databases.

Advanced Build: An MCP-powered business AI agent that securely connects to a mock CRM to fetch and update records.

• LLM Ops fundamentals and lifecycle management.
• Logging, tracing, and debugging with tools like LangSmith or Arize.
• Cost tracking and latency optimization techniques.
• A/B testing for different models and prompt versions.
• Building automated evaluation datasets and regression testing.
• Continuous improvement loops based on user feedback.

Advanced Build: An LLM performance dashboard that tracks token usage, latency, and response quality of an API.

• Understanding and mitigating AI hallucinations.
• Defending against prompt injection and jailbreaking attacks.
• Data privacy, PII redaction, and secure data handling.
• Implementing AI guardrails (e.g., NeMo Guardrails, Llama Guard).
• Bias detection and fairness in AI-generated content.
• Navigating AI compliance (GDPR, EU AI Act, corporate policies).

Advanced Build: A secure GenAI API endpoint protected by input/output guardrails and PII redaction filters.

• Designing scalable, production-ready GenAI system architecture.
• Containerizing AI applications using Docker.
• Serving AI models via FastAPI and deploying to cloud platforms (AWS/GCP).
• CI/CD pipeline basics for AI applications.
• GitHub portfolio best practices and technical documentation.
• Project presentation and technical interview preparation.

Expert Build: An end-to-end, deployed Enterprise Agentic AI Copilot combining LLM, RAG, MCP, and Agents with live documentation.