Name: Generative AI Essentials
Price: 14000 PKR
Availability: InStock

About this course

Hybrid, reranked, agentic RAG in LangChain - from tokens and embeddings to a production-served system.

This is a 6-week short course - enough depth to build real projects and a portfolio piece, without a long commitment. It runs online in the August 2026 cohort (starting 1 August 2026) and is taught the EchoLens way: you learn by doing real, gradeable work rather than just watching lectures.

What's included

Live, instructor-led online sessions across 6 weeks (24 hours total).
Hands-on coding quests you solve inside the EchoLens browser compiler - nothing to install.
Gems, stages and a leaderboard that keep you moving instead of grade anxiety.
A verified certificate with a scannable QR code, ready to share on LinkedIn, when you finish.
The first week is open free so you can try the course before you pay.

What you will learn

Tokenization and LLM internals for buildersChunking and ingestion at engineering qualityEmbeddings and vector index engineeringRAG from baseline to tunedQuery transformation and hybrid searchReciprocal rank fusion and cross-encoder rerankingGraphRAG and agentic RAG with LangGraphFine-tuning with LoRA and QLoRA versus RAGRAG evaluation with RAGAS and LangSmithSecure, cost-aware production serving

Course learning outcomes

By the end of this course, you will be able to:

CLO 1. Explain how an LLM processes tokens, context, and decoding, and apply that understanding to model, cost, and latency decisions.
CLO 2. Engineer a document ingestion pipeline with appropriate parsing, chunking, metadata, and embedding choices.
CLO 3. Build a grounded, cited RAG pipeline in LangChain using LCEL.
CLO 4. Improve retrieval quality through query transformation, hybrid dense and sparse search, and reranking, and justify each choice with metrics.
CLO 5. Design advanced retrieval using GraphRAG and agentic routing with LangGraph.
CLO 6. Evaluate the decision between retrieval and fine-tuning, and apply parameter-efficient fine-tuning with LoRA or QLoRA where it is the right tool.
CLO 7. Evaluate a RAG system with faithfulness and context metrics using RAGAS and LangSmith, and turn the evaluation into a regression test.
CLO 8. Deploy a secure, cost-aware RAG service, addressing prompt injection, guardrails, and PII.

Course outline - level by level

12 leveles, each with hands-on quests you clear in the portal.

Level 1. Tokens and LLM internals · The model landscape and LCEL - Tokenize text and read the context window and its cost, control decoding with temperature and top-p, get oriented in the model landscape from closed to open weights and quantization, and build your first LCEL chain in LangChain.
Level 2. Chunking and document processing · Ingestion pipelines - Parse messy real-world inputs like tables, scanned PDFs and code, compare recursive, token-based, semantic and structure-aware chunking with overlap tuning, design metadata for later filtering, and build an ingestion pipeline that re-indexes incrementally.
Level 3. Embeddings and vector indexes · Recall versus speed - Choose an embedding model by dimensionality, domain fit and MTEB-style benchmarks, understand HNSW, IVF and product quantization across Chroma, FAISS, pgvector and Qdrant, and tune similarity metrics for recall against speed.
Level 4. Building a RAG pipeline with LCEL · Grounded, cited answers - Wire up the retrieve, augment, generate architecture in LCEL, return structured, cited answers validated with Pydantic, and find where a baseline RAG system honestly breaks.
Level 5. Improving retrieval quality · Recall at k and MRR - Transform queries with multi-query and HyDE, route and filter by metadata, retrieve small chunks but return their larger parent context, and measure retrieval with recall at k and mean reciprocal rank.
Level 6. Hybrid RAG · Dense and sparse fusion - Compare BM25 sparse retrieval to dense semantic retrieval and see exactly where each fails, build a hybrid ensemble with alpha weighting and reciprocal rank fusion, and find where hybrid pays off most: names, codes, IDs and rare terms.
Level 7. Reranking · Two-stage retrieval - Cast a wide net for recall and rerank with a cross-encoder for precision through contextual compression, then tune the recall-then-rerank tradeoff against its cost and latency.
Level 8. GraphRAG · Structured retrieval - Handle multi-hop and relationship questions that flat chunks cannot answer by building and querying a knowledge graph alongside vectors, and turn a question into a filtered structured search with self-query retrieval.
Level 9. Agentic RAG with LangGraph · Corrective and self-RAG - Expose RAG as a tool an agent can call, build a LangGraph flow of state, nodes and edges, and add corrective and self-RAG loops that grade retrieved context and verify the final answer.
Level 10. Fine-tuning versus RAG · LoRA and QLoRA - Weigh the real engineering decision between retrieving and fine-tuning, apply parameter-efficient fine-tuning with LoRA and QLoRA, and see what pairing a fine-tuned model with RAG actually buys you.
Level 11. Evaluation and observability · RAGAS and LangSmith - Separate bad retrieval from bad generation with faithfulness, answer relevance and context precision and recall, trace every step in LangSmith, and turn a fixed evaluation set into a regression test.
Level 12. Production, security, and the final project · Serving a real system - Serve behind FastAPI with streaming and caching, run open models locally with Ollama or vLLM, and defend against prompt injection with guardrails and PII handling before shipping the capstone.

How you submit: Coding quests in the built-in compiler with an AI copilot beside the editor - exactly the Cursor/Copilot workflow the course teaches.

Your production end project

End project - your production build

An Advanced, Production-Grade RAG System

A hybrid, reranked, agentic RAG system over a real corpus, evaluated and served - the project that proves you can engineer GenAI, not just call an API.

Build a complete retrieval system over a real document set of your choice, to a standard you could defend in a senior interview. It ingests and indexes the corpus, retrieves with hybrid search and reranking, routes harder questions through an agent, answers with citations, is measured against a baseline, and is served behind an API.

✓A real corpus ingested with engineering-quality parsing, chunking, and metadata
✓A RAG pipeline in LCEL returning grounded, cited, structured answers
✓Hybrid retrieval fusing BM25 and dense search with reciprocal rank fusion
✓A cross-encoder or rerank-service reranking stage applied via contextual compression
✓Either a GraphRAG or an agentic LangGraph layer that routes and runs a corrective or self-RAG grading loop
✓A clear, written justification of RAG versus fine-tuning for this problem
✓An evaluation on a fixed question set reporting faithfulness and context relevance, traced in LangSmith, showing your system beats a plain vector-only baseline
✓Served behind a FastAPI endpoint with streaming, caching, and basic injection guardrails

Shipped when: The system answers real questions over your corpus with correct citations, hybrid retrieval plus reranking and routing measurably outperform the naive baseline on your evaluation set, the whole thing runs behind an API someone else can call, and you can defend every architectural choice.

Who it's for

Generative AI Essentials suits learners at a beginner to intermediate level who want a practical, project-based route into Generative AI Essentials. You need only a browser and an internet connection - all coding runs inside the EchoLens compiler, so there is nothing to set up.

Certificate

Finish every stage and EchoLens issues a verified certificate carrying a QR code anyone can scan to confirm it on our site. You can add it to your CV or share it to LinkedIn in one click.