MLOps
模型部署、评估与运维
deep-learning-core
levy-n
Explains neural network fundamentals: the Three Pillars (Model, Loss, Optimizer), backpropagation, gradient descent variants (SGD, Adam), regularization (Dropout, BatchNorm), and MLP architecture design. Use when learning how neural networks work, debugging training issues, or when user asks about 'backpropagation', 'vanishing gradients', 'learning rate', 'loss function', 'overfitting', 'underfitting', 'activation functions', 'why isn\'t my model learning', 'gradient descent', 'Adam', 'Dropout', 'BatchNorm', 'autoencoder', 'denoising autoencoder', or 'latent space'.
ml-knowledge-index
levy-n
Routes ML/DL questions to specialized skills. Use FIRST when unsure which skill applies, when user asks broad ML questions, or when multiple topics might be relevant. Maps: regression/classification → ml-fundamentals, ensembles/clustering → ml-advanced, TF-IDF/Word2Vec → nlp-classical, training/backprop → deep-learning-core, PyTorch → pytorch-mastery, CNNs/images → cnn-vision, LSTM/time-series → sequence-models, BERT/HuggingFace → transformers-llm, RAG/embeddings → rag-retrieval, APIs/PDF-parsing → data-pipeline, LoRA/QLoRA/PEFT → fine-tuning-peft, MLflow/W&B/Optuna → mlops-experiment, SHAP/Grad-CAM → model-interpretability, Q-learning/PPO/DQN → reinforcement-learning, GAN/VAE/diffusion → generative-models, explanations → ml-teaching-assistant.
agentic-ui-contract
phrazzld
Design and implement agentic product flows using the contract: model decides what to do, tools decide how it is done, UI schema decides how it is rendered. Use for chat-first apps, tool-calling agents, generative UI systems, and planner/tool architecture decisions. Keywords: agentic UX, tool calling, planner, generative UI, function tools.
ml-fundamentals
levy-n
Implements classical ML algorithms for regression and classification. Covers Linear/Polynomial/Logistic Regression, Decision Trees, Ridge/Lasso regularization, train/test splits, cross-validation, and evaluation metrics (R², RMSE, Precision, Recall, F1, ROC-AUC, Confusion Matrix). Use when building predictive models on tabular data, comparing baseline algorithms, handling imbalanced data, or when user mentions 'regression', 'classification', 'overfitting', 'cross-validation', 'confusion matrix', 'feature importance', 'precision/recall', or 'regularization'.
nlp-classical
levy-n
Implements traditional NLP techniques before transformers. Covers text vectorization (TF-IDF, Bag-of-Words), word embeddings (Word2Vec, FastText, GloVe, Doc2Vec), topic modeling (LDA, Gensim), and text similarity (Jaccard, Cosine, FuzzyWuzzy, record linkage). Use when building text classifiers without deep learning, doing topic extraction, entity matching, or when user mentions 'TF-IDF', 'Word2Vec', 'topic modeling', 'LDA', 'text similarity', 'n-grams', 'document clustering', 'GloVe', 'Doc2Vec', 'FuzzyWuzzy', or 'record linkage'.
ml-dl-expert
levy-n
Expert ML/DL teaching assistant for Hebrew University AI Engineering course. Activates for ANY machine learning or deep learning question: neural networks, PyTorch, TensorFlow, transformers, BERT, GPT, RAG, embeddings, CNNs, RNNs, LSTM, NLP, computer vision, clustering, regression, classification, training loops, backpropagation, loss functions, optimization, HuggingFace, vector stores, FAISS, ChromaDB, recommender systems, matrix factorization, transfer learning, data augmentation, autoencoders, Word2Vec, TF-IDF, topic modeling, prompt engineering, fine-tuning, LoRA, QLoRA, PEFT, quantization, sentiment analysis, image classification, object detection, time series, XGBoost, Random Forest, PCA, t-SNE, DBSCAN, K-Means, data pipeline, PDF parsing, chunking, function calling, AI agents, MLflow, W&B, experiment tracking, hyperparameter tuning, Optuna, SHAP, feature importance, Grad-CAM, model interpretability, reinforcement learning, Q-learning, DQN, PPO, policy gradient, GANs, VAE, diffusion models, Stable Diffusion, generative AI, model deployment, MLOps, synthetic data, data sourcing, Kaggle, dataset, data augmentation, SMOTE. Routes to 17 specialized sub-skills and provides code examples, visual diagrams, and Hebrew explanations when needed.
Model Patterns
Kaakati
"Entity and model patterns with JSON serialization, immutability, and equality"
reinforcement-learning
levy-n
Reinforcement learning fundamentals and practical implementations. Covers RL concepts (agent, environment, reward, policy), Q-Learning, Deep Q-Network (DQN), Policy Gradient methods, PPO, Actor-Critic, Gymnasium environments, Stable-Baselines3, reward shaping, and exploration-exploitation trade-off. Use when user asks about 'reinforcement learning', 'RL', 'Q-learning', 'DQN', 'PPO', 'policy gradient', 'reward function', 'agent', 'environment', 'Gym', 'Gymnasium', 'exploration', 'exploitation', 'Stable-Baselines3', 'Actor-Critic', 'SARSA', 'Bellman equation', or 'Markov decision process'.
ml-advanced
levy-n
Implements ensemble learning (Random Forest, XGBoost, CatBoost, Stacking) and unsupervised methods (K-Means, DBSCAN, Hierarchical clustering, PCA, t-SNE, UMAP), and recommender systems (Matrix Factorization, NeuMF). Use when comparing gradient boosting algorithms, doing customer segmentation, anomaly detection, dimensionality reduction, building recommender systems, or when user mentions 'ensemble', 'boosting', 'bagging', 'random forest', 'XGBoost', 'clustering', 'K-Means', 'DBSCAN', 'elbow method', 'silhouette score', 'PCA', 't-SNE', 'dimensionality reduction', 'feature importance', 'matrix factorization', 'NeuMF', 'recommender system', or 'collaborative filtering'.
ml-teaching-assistant
levy-n
Explains ML/DL concepts with analogies, visual diagrams, and progressive complexity. Covers backpropagation, gradient descent, attention mechanisms, neural networks, ML project methodology, and 50+ other concepts. Also provides the 5-step ML workflow, anti-patterns checklist, and model selection decision trees. Use when user says 'explain', 'I don\'t understand', 'how does X work', 'teach me', 'why does', 'what is the intuition', 'how should I approach', 'best practice', 'common mistakes', 'workflow', 'methodology', or asks conceptual 'why' questions about any ML topic. Provides intuitive explanations before math, ASCII visualizations, everyday analogies, and corrects common misconceptions.
ml-engineering
eyadsibai
Use when "deploying ML models", "MLOps", "model serving", "feature stores", "model monitoring", or asking about "PyTorch deployment", "TensorFlow production", "RAG systems", "LLM integration", "ML infrastructure"
video-agent
founderjourney
AI content generation suite with 35+ models. Image generation, video creation, audio processing via FAL AI, Google Vertex AI, ElevenLabs. Pipeline orchestration and cost management.
Repository Patterns
Kaakati
"Repository interface and implementation patterns with offline-first strategies"
modelslab-model-discovery
ModelsLab
Search and discover 50,000+ AI models on ModelsLab, check usage analytics, and monitor generation history via the Agent Control Plane API.
Error Handling Patterns
Kaakati
"Exception classes, failure classes, Either type, and error handling strategies"
ActiveRecord Query Patterns
Kaakati
"Complete guide to ActiveRecord query optimization, associations, scopes, and PostgreSQL-specific patterns. Use this skill when writing database queries, designing model associations, creating migrations, optimizing query performance, or debugging N+1 queries and grouping errors."
modal
eyadsibai
Use when "Modal", "serverless GPU", "cloud GPU", "deploy ML model", or asking about "serverless containers", "GPU compute", "batch processing", "scheduled jobs", "autoscaling ML"
gemini-api
diskd-ai
Google Gemini API integration for building AI-powered applications. Use when working with Google's Gemini API, Python SDK (google-genai), TypeScript SDK (@google/genai), multimodal inputs (image, video, audio, PDF), thinking/reasoning features, streaming responses, structured outputs with JSON schemas, multi-turn chat, system instructions, image generation (Nano Banana), video generation (Veo), music generation (Lyria), embeddings, document/PDF processing, or any Gemini API integration task. Triggers on mentions of Gemini, Gemini 3, Gemini 2.5, Google AI, Nano Banana, Veo, Lyria, google-genai, or @google/genai SDK usage.
prompt-engineering
maragudk
"Use this skill when crafting, reviewing, or improving prompts for LLM pipelines — including task prompts, system prompts, and LLM-as-Judge prompts. Triggers include: requests to write or refine a prompt, diagnose why an LLM produces inconsistent or incorrect outputs, bridge the gap between intent and model behavior, reduce ambiguity in instructions, add few-shot examples, structure complex prompts, or improve output formatting. Also use when the user needs help distinguishing specification failures (unclear instructions) from generalization failures (model limitations), or when iterating on prompts based on observed failure modes. Do NOT use for general coding tasks, document creation, or non-LLM writing."
mlops-experiment
levy-n
MLOps and experiment tracking for reproducible ML workflows. Covers MLflow, Weights & Biases (W&B), TensorBoard, hyperparameter tuning (Optuna, Ray Tune), model registry, experiment versioning, and production deployment patterns. Use when user asks about 'MLflow', 'W&B', 'Weights and Biases', 'experiment tracking', 'hyperparameter tuning', 'Optuna', 'model registry', 'TensorBoard', 'reproducibility', 'model versioning', 'ML pipeline', 'model deployment', 'logging', 'wandb', or 'Ray Tune'.
announce
phrazzld
Launch post generator for multiple platforms. Creates Twitter, HN, Reddit, and Indie Hackers announcements. Use for product launches, major updates, or milestone announcements.
pytorch-mastery
levy-n
Implements PyTorch training patterns, data loading, and GPU optimization. Covers tensor operations, DataLoader/Dataset classes, training loops, CUDA memory management, and debugging common errors. Use when writing PyTorch code, debugging tensor shape mismatches, fixing CUDA OOM errors, optimizing training speed, or when user mentions 'PyTorch', 'tensor', 'DataLoader', 'training loop', 'GPU memory', 'CUDA', '.to(device)', 'model.eval()', 'torch.no_grad()', 'shape mismatch', 'environment setup', 'nvidia-smi', or 'CUDA setup'.
notion-mastery
founderjourney
Sistema completo de productividad y CRM en Notion con integracion n8n. Usar cuando el usuario necesite gestionar tareas, proyectos, metas, pipeline de ventas, CRM de clientes, prospeccion con Apollo.io, o automatizar workflows entre Notion y otras herramientas. Activa con palabras como Notion, tareas, proyectos, CRM, pipeline, leads, Apollo, prospeccion, follow-up, deals, clientes.
workout-program-designer
10x-Anit
Custom training plans by goal (strength, cardio, flexibility). Progressive overload programming, rest day optimization, home vs gym adaptations, deload weeks.