AI Engineering Training in Perungudi

• Advanced Python (Asyncio, decorators, type hinting)
• Building robust REST and GraphQL APIs with FastAPI
• Containerization with Docker for AI applications
• Basic CI/CD pipelines for AI code deployment
• Cloud fundamentals for AI (AWS/GCP basics)

Beginner to Intermediate Build: Deploy an asynchronous FastAPI data ingestion pipeline containerized with Docker.

• Transformer architecture and attention mechanisms (conceptual)
• Tokenization, embeddings, and context window management
• Advanced prompting: Chain-of-Thought (CoT) and ReAct
• Structured output generation (JSON mode, function calling)
• Prompt injection vulnerabilities and defensive prompting

Beginner Build: Create an intelligent Prompt Optimizer tool that refines vague inputs into structured prompts.

• Deep dive into text, image, and audio embeddings
• Setting up and querying modern Vector DBs (Qdrant, Pinecone)
• Hybrid search (combining dense vectors with sparse BM25)
• Metadata filtering and namespace management
• Implementing cross-encoder reranking for higher accuracy

Beginner to Intermediate Build: Develop a multi-modal semantic search engine ranking text and images via hybrid retrieval.

• Advanced chunking strategies (semantic, document-aware)
• Introduction to GraphRAG and knowledge graph integration
• Agentic RAG: Autonomous query routing and self-correction
• Contextual retrieval and handling multi-modal documents
• Evaluating RAG pipelines for faithfulness and relevance

Advanced Build: Deploy an enterprise GraphRAG assistant that answers analytical questions from complex PDFs with precise citations.

• The anatomy of an AI Agent (Perception, Reasoning, Action)
• Deep dive into Model Context Protocol (MCP) architecture
• Building custom MCP servers and clients for enterprise tools
• Tool selection, execution, and error handling loops
• Memory management (short-term vs. long-term agent memory)

Advanced Build: Create an MCP-powered agent that autonomously updates CRM, Slack, and Jira tickets securely.

• Frameworks for multi-agent systems (LangGraph, CrewAI)
• Designing agent personas, roles, and backstories
• State management and shared memory across agents
• Human-in-the-loop (HITL) workflows and approval gates
• Handling agent deadlocks and infinite execution loops

Advanced Build: Orchestrate a multi-agent team to autonomously research, analyze data, and generate comprehensive industry reports.

• When to fine-tune vs. when to use RAG/Prompting
• Parameter-Efficient Fine-Tuning (LoRA, QLoRA)
• Direct Preference Optimization (DPO) for alignment
• Training and evaluating SLMs (Llama-3, Phi-4, Gemma)
• Quantization techniques (GGUF, AWQ) for edge deployment

Advanced Build: Fine-tune a 7B SLM on a specialized dataset using DPO and deploy it locally.

• Defining evaluation metrics (Faithfulness, Relevancy, Toxicity)
• "LLM-as-a-Judge" frameworks and pairwise evaluation
• Building automated regression test suites for LLMs
• Implementing input/output guardrails (NeMo Guardrails)
• PII detection and data masking in AI pipelines

Intermediate to Advanced Build: Create a CI/CD pipeline evaluating RAG systems against golden datasets to block inaccurate deployments.

• Serving LLMs at scale (vLLM, SGLang, TensorRT-LLM)
• Observability and tracing (LangSmith, Arize Phoenix)
• Caching strategies (Semantic caching) to reduce API costs
• Blue-green deployments and A/B testing for AI models
• Serverless GPU architectures and cost optimization

Advanced Build: Deploy a highly concurrent LLM API using vLLM, semantic caching, and real-time observability dashboards.

• Native multimodal models (processing image/audio natively)
• Voice AI pipelines (STT, LLM, TTS) with ultra-low latency
• WebRTC integration for real-time conversational AI
• Running multimodal models on edge devices (ONNX)
• Optimizing bandwidth and compute for edge deployments

Advanced Build: Develop a real-time, voice-activated multimodal assistant that analyzes webcam input to answer environmental questions.

• AI Red-teaming and adversarial attack simulations
• Model extraction, data poisoning, and defense strategies
• Compliance with the EU AI Act and global AI regulations
• Implementing audit trails and explainability (XAI)
• Ethical AI frameworks and bias mitigation techniques

Intermediate to Advanced Build: Conduct a security audit and red-team simulation on an AI agent, implementing automated guardrails.

• System design for complex, scalable AI architectures
• Writing technical documentation and API specifications
• Resume building and portfolio optimization for AI roles
• Mock technical interviews (System design + Coding)
• Navigating the AI job market and freelance opportunities

Expert Build: Deploy a production-ready AI SaaS application from scratch, complete with CI/CD, monitoring, and frontend.