Browse Skills

Expert guide for deploying TensorFlow models to production using TensorFlow Serving, covering SavedModel optimization, serving infrastructure, batching strategies, model versioning, and monitoring for reliable ML inference at scale.

Evaluates machine learning models with rigorous methodology and statistical analysis. Covers metrics selection, cross-validation strategies, bias detection, A/B testing, confusion matrices, ROC/AUC analysis, and comprehensive evaluation reporting.

Provides expert guidance on time series forecasting including statistical methods (ARIMA, ETS), Prophet, neural forecasting (N-BEATS, Temporal Fusion Transformer), seasonality decomposition, anomaly detection, and confidence interval estimation.

Guide the end-to-end curation of high-quality datasets for machine learning. Covers data collection strategies, cleaning pipelines, annotation workflows, bias auditing, deduplication, and dataset documentation following data-centric AI principles.

Designs optimal text chunking strategies for Retrieval-Augmented Generation (RAG) systems, balancing chunk size, semantic coherence, retrieval precision, and generation quality across diverse document types.

Design, tune, and evaluate search systems for maximum relevance using BM25, hybrid retrieval, learning-to-rank, and neural re-ranking with rigorous offline and online evaluation frameworks.

Optimizes ML model deployment for production inference including quantization (INT8/INT4), knowledge distillation, ONNX/TensorRT conversion, model pruning, batching strategies, and edge deployment for reducing latency and cost.

Designs comprehensive evaluation frameworks for LLM and AI systems, combining automated metrics (BLEU, ROUGE, BERTScore), LLM-as-judge patterns (G-Eval), RAG evaluation (RAGAS), standard benchmarks (MMLU, HumanEval), safety evaluations, and A/B testing methodologies into production-ready evaluation pipelines.

Design and optimize manufacturing facility layouts using digital twin simulation. Create virtual replicas of production floors to test layout alternatives, identify bottlenecks, and validate material flow improvements before physical implementation.

Designs privacy-preserving distributed machine learning systems using federated learning. Covers federated averaging, differential privacy, secure aggregation, communication efficiency, and compliance with privacy regulations.

Performance tuning for Pinecone, Weaviate, Qdrant, and pgvector including index selection, embedding dimensionality, chunking strategies, and hybrid search optimization for RAG systems.

Designs and optimizes retrieval-augmented generation systems for customer support operations, including knowledge base ingestion, chunking strategies, hybrid retrieval pipelines, answer generation with citation, and continuous quality improvement loops.

Designs pedagogically sound automated feedback systems for programming education. Creates error taxonomies, scaffolded hint systems, and feedback message architectures that transform coding errors into effective learning opportunities.

Design comprehensive evaluation systems for LLMs and AI applications, covering automated metrics, human evaluation protocols, LLM-as-a-judge pipelines, and CI/CD-integrated regression testing.

Design and implement machine learning-based lead scoring systems that predict conversion probability, prioritize sales efforts, and automate lead qualification using behavioral, firmographic, and engagement data from CRM platforms.

Guides the end-to-end deployment of TensorFlow Lite models on mobile and edge devices — from model conversion and quantization to on-device inference optimization, hardware delegate selection, and production monitoring. Ensures models meet latency, size, and accuracy constraints for resource-constrained environments.

Expert in unsupervised learning techniques — clustering, dimensionality reduction, anomaly detection, and pattern discovery with scikit-learn, UMAP, and HDBSCAN.

Design synthetic data generation pipelines that produce privacy-preserving, statistically faithful datasets for ML training, testing, and data sharing using GANs, copulas, and differential privacy.

Designs comprehensive data labeling strategies including annotation pipeline architecture, inter-annotator agreement measurement, active learning loops, label quality control, crowdsourcing management, few-shot labeling, and weak supervision techniques.

Reduces LLM API costs through systematic token optimization — prompt compression, caching strategies, model routing, and batch processing — while maintaining output quality across OpenAI, Anthropic, and open-source models.

Designs, builds, and validates high-quality training datasets for LoRA fine-tuning of large language models, applying data-centric AI principles and instruction-tuning best practices to maximize model performance per training example.