mlx

Builds AI-powered applications using pre-trained models, LLM APIs, embeddings, RAG pipelines, and agent architectures. Knows the Claude Agent SDK, OpenAI Agents SDK, Vercel AI SDK, and DSPy — and fetches their live docs before scaffolding agent code. Use proactively when the user wants to build an AI application, set up a RAG system, do prompt engineering, integrate LLM APIs, build an agent with any framework, work with embeddings/vector stores, optimize prompts with DSPy, or evaluate LLM outputs.

Answers business questions with data through descriptive statistics, hypothesis testing, segmentation, trend analysis, and visualization. Use proactively when the user wants to understand what happened in the data, compare groups, find trends, create charts or dashboards, run A/B test analysis, segment customers, calculate KPIs, or build data reports for stakeholders.

Builds and maintains data pipelines, warehouses, and lakehouses. Use proactively when the user needs to build an ETL/ELT pipeline, set up dbt transformations, implement incremental loading, orchestrate workflows with Airflow or Prefect, process data at scale with Spark, Polars, or DuckDB, design a data lakehouse (Delta Lake, Iceberg, Hudi), validate data quality with Great Expectations or Soda, or set up production data infrastructure feeding ML systems. Distinct from data-scientist (exploratory modeling) and data-analyst (BI and reporting).

Full-pipeline data science agent: dataset discovery, EDA, cleaning, feature engineering, training, and evaluation. Use proactively when the user needs the COMPLETE workflow from finding data to trained model, or has a dataset and needs exploration through modeling. Always starts with data understanding.

Specializes in neural network architecture design, training dynamics, and GPU optimization. Use proactively when the user needs to design or debug a neural network architecture (CNNs, Transformers, RNNs, SSMs, diffusion models), troubleshoot loss curves or gradient pathologies (vanishing gradients, exploding gradients, dead ReLU), implement distributed training (DDP, FSDP, DeepSpeed), optimize GPU memory and throughput (mixed precision, gradient checkpointing, torch.compile), or run systematic architecture search experiments. Distinct from ml-engineer (tabular/classical ML) and ai-engineer (pre-trained model integration).

Statistical analysis, hypothesis testing, A/B testing, cohort analysis, segmentation, trend detection, business metrics, pre-delivery validation, and data visualization. Use when the user asks to "analyze this data", "run a statistical test", "compare groups", "find trends", "do A/B test analysis", "segment customers", "calculate KPIs", "validate this analysis", "check my work", "sanity check", "review my numbers", "make a chart", "create a dashboard", "plot the data", "visualize results", or mentions hypothesis testing, cohort analysis, business analytics, data validation, bar charts, line charts, heatmaps, scatter plots, or data storytelling.

Autonomous time-budget experiment loop. Modify a training script, train for a fixed wall-clock budget, evaluate, record, repeat. Inspired by karpathy/autoresearch. Use for overnight architecture search, systematic hyperparameter sweeps, or any iterative model improvement workflow.

Context engineering for building production LLM applications: context window management, degradation patterns, optimization strategies, memory system selection, multi-agent architecture, filesystem context patterns, and tool design principles. Use when building LLM apps, RAG pipelines, AI agents, multi-agent systems, or when designing memory, tool APIs, or context strategies for any language model application.

Explore, clean, and engineer datasets end-to-end: statistical profiling, distribution checks, missing value analysis, duplicate detection, outlier removal, type fixing, encoding, create features, encode categories, transform columns, add rolling windows, build interaction terms, and feature engineering. Supports pandas, polars, and PySpark. Use when the user wants to explore data, profile columns, understand a dataset, clean data, handle missing values, remove duplicates, fix data types, preprocess a dataset before modeling, create features, encode categories, transform columns, add rolling windows, build interaction terms, or do feature engineering.

Detect data drift, concept drift, and model performance degradation in production. Uses PSI, KS-test, and chi-squared for statistical drift, plus evidently and nannyml for automated reports. Use when monitoring a deployed model or comparing training vs production data distributions.

23 hooks across 17 events

pyright-langserver