Lessons
view all ⭢
lesson
Evaluating N-Gram ModelsAI BootcampEvaluate model quality using entropy, character diversity, and negative log likelihood (NLL).
lesson
Building and Sampling N-Gram ModelsAI BootcampConstruct frequency dictionaries, normalize into probability matrices, and sample random text using bigram/trigram models.
lesson
Introduction to N-Gram ModelsAI BootcampUnderstand n-grams and their use in modeling language with simple probabilities, implement bigram/trigram extraction with sliding windows.
lesson
RAG Evaluation & ImplementationAI BootcampEvaluate RAG with `recall@k`, `precision@k`, `MRR`, generate synthetic data with LLMs, and implement baseline vector search with LanceDB and OpenAI embeddings.
lesson
Vector Databases and Query OptimizationAI BootcampExplore vector databases, chunking, and query optimization (HyDE, reranking, filtering) using contrastive learning and cosine similarity.
lesson
RAG Pipeline OverviewAI BootcampUnderstand the RAG pipeline (pre-retrieval, retrieval, post-retrieval) and compare term-based vs embedding-based retrieval (TF-IDF, BM25 vs vector search).
lesson
Advanced Multimodal TasksAI BootcampExperiment with visual question answering (VQA), image captioning, and compare multimodal architectures (CLIP, ViLT, ViT-GPT2).
lesson
Prompt Engineering with ImagesAI BootcampPractice prompt engineering with images, explore wording’s impact on retrieval, and build text-to-image/image-to-image retrieval systems using cosine similarity.
lesson
Introduction to CLIP and Multimodal EmbeddingsAI BootcampUnderstand CLIP’s joint image-text representations via contrastive learning and run similarity queries in shared embedding space.
lesson
Multimodal Embeddings for RetrievalAI BootcampConduct retrieval and classification tasks with image/audio embeddings (CLIP, Wav2Vec2) and explore emerging architectures like Byte Latent Transformers.
lesson
Working with EmbeddingsAI BootcampLearn how embedding vectors represent meaning, manipulate Word2Vec-style embeddings (vector math, dot product similarity), and apply to multimodal models (CLIP, ViLT, ViT-GPT2).
