Lab 1 – Tokenization & Embeddings
Understand how language models interpret text by converting sentences into tokens and embeddings, and solve similarity matching using vector distance.

Lab 2 – LLM Parameters & Behavior Control
Control and optimize LLM outputs by tuning generation parameters such as temperature, top-p, max tokens, and penalties.

Lab 3 – Working with LLM APIs
Integrate an LLM into an application by making API calls, parsing JSON responses, and programmatically extracting outputs.

Lab 4 – Context-Aware Chatbot
Build a multi-turn conversational chatbot that maintains context using role-based message history.

Lab 5 – Prompt Chaining
Design a multi-step reasoning workflow by chaining prompts using LangChain to solve complex tasks or basic RAG scenarios.

Lab 6 – Structured & Reliable Outputs
Generate predictable, schema-validated structured responses using LangChain with JSON or Pydantic models.

Lab 7 – Quantitative Model Evaluation
Evaluate model predictions by calculating precision, recall, and F1 score on a sample classification problem.

Lab 8 – Qualitative LLM Evaluation
Assess and compare LLM responses using custom rubric-based criteria such as relevance, clarity, completeness, and tone.

Lab 9 – Fine-Tuning for Classification
Improve task-specific performance by fine-tuning an LLM on labeled text data and validating it on unseen inputs.

Lab 10 – Parameter-Efficient Fine-Tuning (PEFT)
Optimize fine-tuning cost and speed by training only a subset of model parameters and comparing results with the base model.

Lab 11 – Retrieval-Augmented Generation (RAG)
Build a document-aware GenAI system using LlamaIndex to answer queries grounded in retrieved content.

Lab 12 – Multi-Query RAG Optimization
Improve retrieval accuracy by generating multiple query variations and comparing them against single-query RAG pipelines.

Lab 13 – Agentic Routing
Design an intent-based agent that dynamically selects the correct tool or workflow based on user queries.

Lab 14 – ReAct Agent Implementation
Build a ReAct-style agent that reasons, takes tool actions, observes results, and produces grounded final answers.