Module 1: LLM Foundation
Learn how Large Language Models work from the ground up: tokens, context windows, hyperparameters, prompts, structured outputs, embeddings, semantic search, evaluation, and attention mechanics.
Syllabus Modules
Understand character mappings, subwords vocabulary splits, BPE encoding steps, and API cost implications.
Manage context window capacities, chat history trimming, sliding window states, and RAG query packing.
Deconstruct LLM decoding logic. Explore Temperature, Softmax distribution curves, and penalties.
Master prompt design topologies: system parameters, classifications, injection protections, and few-shots.
Enforce schema structures on unstructured completions using JSON validation frameworks.
Scale ingestion pipelines. Manage batch loops, concurrency pipelines, and rate-limiting limits.
Convert textual characters into high-dimensional vectors to measure similarities mathematically.
Manage database indexing, approximate nearest neighbor algorithms, and metadata search filters.
Deconstruct dot-product attention steps, QKV matrices, and context calculations mathematically.
Decode transformer architecture blocks. Study layer normalizations and feed-forward neural layers.
Design diagnostic evaluation metrics checking hallucination counts, faithfulness, and CI/CD validation checks.
Learning Outcomes
- Master prompt engineering design methodologies
- Deconstruct subword tokenizers and BPE algorithms
- Manage token budgets and context window limits
- Enforce JSON schemas and type-safe structured outputs
- Generate embeddings and perform vector search lookups
- Run evaluation tests using golden sets and LLM-as-a-judge patterns
Interview Defense
- Explain time-space costs of tokenizer inflation
- Defend prompt classification vs fine-tuning strategies
- Analyze context scaling tradeoffs in multi-turn conversations