Machine Learning & AI Topics
Production machine learning systems, model development, deployment, and operationalization. Covers ML architecture, model training and serving infrastructure, ML platform design, responsible AI practices, and integration of ML capabilities into products. Excludes research-focused ML innovations and academic contributions (see Research & Academic Leadership for publication and research contributions). Emphasizes applied ML engineering at scale and operational considerations for ML systems in production.
Generative AI and Large Language Models
The capabilities and behavior of modern generative and large language models. Covers how LLMs are pretrained, in-context learning and few-shot prompting, generative model families (autoregressive, diffusion), context windows, and tokenization and sampling. Emphasizes understanding what generative models can and cannot do and how they differ from discriminative ML.
LLM Evaluation and Observability
Measuring, monitoring and releasing generative and LLM-powered systems. Covers evaluation of open-ended outputs (human review, calibrated model-graded judges, reference-based and embedding scores, perplexity), hallucination, toxicity, privacy-leak and red-team checks, offline benchmarks versus online monitoring, and production observability (logs, traces, metrics, cost and token attribution, privacy-safe telemetry, drift and quality alerts). Also release decisions: A/B tests, canary rollouts, SLOs and go or no-go gates. Building the grounding architecture itself, calibration of predictive models and general software testing are covered elsewhere.
Prompt Engineering
Designing prompts and instructions that steer large language models toward reliable, controllable outputs without changing model weights. Covers zero- and few-shot prompting and example selection, chain-of-thought and step-by-step reasoning prompts, system, developer and role instructions and instruction hierarchy, output formatting and structured output (JSON schemas, templates, constraints), task decomposition and prompt chaining, managing context and token budgets inside a prompt, prompt injection as a design constraint, and iterating on, versioning and regression-testing prompts against a test set. Also covers prompt-level ways to reduce hallucination (grounding in supplied text, quoting support, allowing abstention) and unsafe, toxic or off-policy output (scoped roles, refusal templates), and prompting a model to generate constrained, validated output such as SQL. Boundary: excludes how models are pretrained and behave internally, sampling and decoding internals, fine-tuning, retrieval pipelines, and general LLM evaluation and monitoring.
Recommendation, Ranking, and Personalization
Systems that select and order items for users. Covers candidate generation and ranking (two-stage retrieval, embedding-based retrieval with two-tower models and approximate nearest-neighbour candidate lookup, multi-stage scoring and cross-encoder re-ranking), collaborative and content-based filtering, matrix factorization (alternating least squares, Bayesian personalized ranking), sequence-aware and graph-based recommenders, cold start, learning-to-rank and negative sampling, turning noisy implicit feedback into labels, position, exposure and popularity bias and feedback loops, diversity, fairness-aware exposure and multi-objective re-ranking, counterfactual (off-policy) evaluation of recommenders (inverse propensity scoring, doubly robust estimators), and exploration with multi-armed and contextual bandits (epsilon-greedy, upper-confidence-bound and Thompson sampling, LinUCB, regret, non-stationarity, constraints, rollout, and when a bandit beats a fixed split or full reinforcement learning). Includes end-to-end recommender and personalized feed design. Focuses on the modeling and evaluation patterns specific to recommendation and ranking at scale. General A/B test design, generic ranking-metric definitions and ML serving infrastructure are covered elsewhere.
Retrieval-Augmented Generation (RAG)
Grounding language-model outputs in external knowledge at inference time. Covers document ingestion, chunking and embedding, sparse, dense and hybrid retrieval (BM25, score fusion, contrastive training of retrievers), vector search and approximate-nearest-neighbour index engineering (HNSW, IVF, product quantization, filtering, updates, sharding), re-ranking with cross-encoders, and context assembly with prompt templates and citations. Also covers reducing hallucination through verification and abstention, evaluating retrieval-augmented systems, and production concerns such as freshness, access control, prompt injection through retrieved documents, deletion and privacy, caching, cost, and weighing retrieval against fine-tuning. Generic LLM evaluation and release gates, the craft of prompts beyond retrieved-context templates, attention internals, parameter-efficient fine-tuning, model serving and distributed inference engineering, database indexes, and whole-product ML system design are covered elsewhere.
Transformers and Attention
The transformer architecture behind modern language models. Covers queries, keys and values, scaled dot-product and multi-head attention (including why scores are scaled and worked numeric examples), masking, positional encoding (sinusoidal, learned, relative with bucketed biases, rotary, and 2D schemes for image patches), encoder-only, decoder-only and encoder-decoder variants, residual connections, layer normalization placement and the feed-forward block, parameter, memory and compute accounting, why transformers scale better than recurrent models, efficient and long-context attention variants (sparse, local window with global tokens, hashing-based, linear, low-rank, FlashAttention, multi-query), and what attention heads learn and what raw attention weights do and do not explain. Includes implementing attention and its components from scratch. Distributed training, serving and quantization, fine-tuning, pretraining objectives, and general optimizer practice are covered elsewhere.