levy-n
@levy-n
Public Skills
gsd-orchestration
by levy-n
Spec-driven development orchestration with context engineering for solo developers. Prevents context rot through fresh subagent contexts and atomic task execution. Use when: starting projects, planning features, executing development phases, or when user says "gsd", "plan", "new project", "execute phase". Triggers: /gsd init, /gsd plan, /gsd execute, /gsd status, /gsd verify
cnn-vision
by levy-n
Implements CNN architectures for computer vision tasks. Covers convolution operations, pooling, CNN design patterns (LeNet, ResNet, VGG), transfer learning, fine-tuning pretrained models, data augmentation, and image preprocessing. Use when building image classifiers, doing object detection, or when user mentions 'CNN', 'convolution', 'pooling', 'ResNet', 'VGG', 'transfer learning', 'fine-tuning', 'image augmentation', 'ImageNet', 'feature maps', 'MNIST', 'image classification', 'multi-modal', 'image captioning', or 'multimodal network'.
ml-advanced
by 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'.
agent-architect
by levy-n
Builds autonomous AI agent systems using Claude Agent SDK. Works with Claude MAX subscription - no API keys required. Conducts an architectural interview, proposes architecture, and after approval — builds everything. Covers subagents, MCP tools, hooks, orchestration, and production deployment.
model-interpretability
by levy-n
Model interpretability, explainability, and debugging tools. Covers SHAP (TreeExplainer, DeepExplainer, KernelExplainer), feature importance analysis, LIME, attention visualization, Grad-CAM for CNNs, confusion matrix analysis, error analysis patterns, and model fairness auditing. Use when user asks about 'SHAP', 'feature importance', 'explainability', 'interpretability', 'why did the model predict', 'Grad-CAM', 'LIME', 'attention weights', 'confusion matrix', 'error analysis', 'model debugging', 'fairness', 'bias detection', or 'what did the model learn'.
rag-retrieval
by levy-n
Implements RAG (Retrieval-Augmented Generation) pipelines. Covers embedding APIs (OpenAI, Gemini, Sentence-Transformers), vector stores (FAISS, ChromaDB, Pinecone), RAG variants (Query Rewrite, Conversational, Multi-hop), and evaluation (RAGAS, Faithfulness). Use when building knowledge bases, semantic search, chatbots with documents, or when user mentions 'RAG', 'embeddings', 'vector store', 'FAISS', 'ChromaDB', 'similarity search', 'retrieval', 'chunking', 'hallucination reduction', 'semantic search', or 'knowledge base'.
data-pipeline
by levy-n
Data ingestion pipelines, LLM APIs, document processing, and ML data sourcing. Covers LLM API setup (OpenAI, Gemini, Ollama, LiteLLM), PDF/HTML extraction, chunking, function calling, structured output, dataset sourcing, synthetic data generation, and augmentation. Keywords: 'PDF extraction', 'chunking', 'LiteLLM', 'function calling', 'API setup', 'OCR', 'Pydantic', 'data ingestion', 'dataset', 'Kaggle', 'HuggingFace datasets', 'synthetic data', 'SMOTE', 'augmentation', 'data sourcing', 'active learning'.
ml-fundamentals
by 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'.
transformers-llm
by levy-n
Implements Transformer models and LLM workflows. Covers attention mechanism, BERT fine-tuning, HuggingFace Transformers library (Tokenizer, Trainer, Pipeline), and LLM ecosystem (GPT, Claude, Gemini, Ollama). Use when fine-tuning language models, using HuggingFace, calling LLM APIs, or when user mentions 'transformer', 'attention', 'BERT', 'HuggingFace', 'tokenizer', 'fine-tuning', 'LLM', 'GPT', 'Claude', 'Gemini', 'prompt engineering', 'zero-shot', or 'few-shot learning'.
deep-learning-core
by 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'.
nlp-classical
by 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-knowledge-index
by 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.
reinforcement-learning
by 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-teaching-assistant
by 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.
fine-tuning-peft
by levy-n
Expert guide for LLM fine-tuning and parameter-efficient training methods. Covers LoRA, QLoRA, PEFT library, adapter tuning, instruction tuning, quantization (GPTQ, AWQ, GGUF, bitsandbytes), dataset preparation for fine-tuning, Hugging Face Trainer/TRL, RLHF/DPO/ORPO alignment, and model merging. Use when user asks about 'fine-tuning', 'LoRA', 'QLoRA', 'PEFT', 'adapter', 'quantization', 'bitsandbytes', '4-bit', '8-bit', 'instruction tuning', 'RLHF', 'DPO', 'model merging', 'Unsloth', 'Axolotl', 'training custom models', 'TRL', or 'SFT'.
pytorch-mastery
by 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'.
ml-dl-expert
by 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.
mlops-experiment
by 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'.
generative-models
by levy-n
Generative AI models: GANs, VAEs, Diffusion Models, and image generation. Covers GAN architecture (Generator/Discriminator), DCGAN, Wasserstein GAN, Variational Autoencoders, latent space interpolation, Diffusion models (DDPM), Stable Diffusion, conditional generation, and text-to-image. Use when user asks about 'GAN', 'generative adversarial', 'VAE', 'variational autoencoder', 'diffusion model', 'image generation', 'Stable Diffusion', 'DCGAN', 'Wasserstein', 'WGAN', 'latent space', 'generate images', 'text-to-image', 'DDPM', 'denoising diffusion', 'style transfer', or 'deepfake'.
sequence-models
by levy-n
Implements sequence models for time series and text. Covers RNN fundamentals, LSTM/GRU architectures, time series forecasting, text generation with language models, and sequence classification. Use when working with sequential data, predicting time series, text generation, or when user mentions 'RNN', 'LSTM', 'GRU', 'vanishing gradient', 'hidden state', 'time series', 'sequence-to-sequence', 'text generation', 'next word prediction', or 'recurrent neural network'.
prompt-master
by levy-n
Expert prompt engineer that creates perfect, optimized prompts for any task. Conducts a structured AskUserQuestion interview, scans available skills for integration, selects the right techniques, and builds Claude-optimized prompts. Use when user says "create a prompt", "build me a prompt", "I need a prompt for", "optimize this prompt", "prompt master", or needs help engineering any prompt.
SVG Logo Designer
by levy-n
"Create professional SVG logos from descriptions and design specifications. Generates multiple logo variations with different layouts, styles, and concepts. Produces scalable vector graphics that can be used directly or exported to PNG. Use this skill when users ask to create logos, brand identities, icons, or visual marks for their designs."