3. Data, Science & AI

Generates images from text prompts using Google Gemini AI, serving as a backend for image-related skills like cover-image and article-illustrator.

Generates images from text or reference using OpenAI and Google AI APIs with customizable aspect ratios and quality settings.

Compresses images to WebP by default with PNG-to-PNG support using system tools and Sharp fallback.

Assists in crafting and running SQL queries, from basic single-table operations to advanced multi-table joins and aggregations.

Explores and analyzes database schemas to understand table structures, columns, and relationships.

Generates data analysis reports with SQL queries, statistical insights, and interactive HTML visualizations based on user requirements.

Ensures data assets have accurate and complete metadata for proper handling, searchability, and integration in data pipelines.

Debugs MLX model ports by comparing exported tensors and images against PyTorch/diffusers reference implementations.

Ports ML models to mflux/MLX with correctness validation and style refactoring for AI frameworks.

Enables unified coordination of multiple AI agents to collaboratively solve complex tasks through structured orchestration.

Guides PyTorch-native reinforcement learning training with Meta's torchforge library for clean abstractions and scalable experiments using Monarch and TorchTitan.

Guides LLM post-training with reinforcement learning (RLHF, PPO) using verl (Volcano Engine RL) for scalable infrastructure-backed implementation.

Guides LLM post-training with RL using slime framework (Megatron+SGLang) for GLM models and custom data workflows.

Guides enterprise RL training with miles (slime fork) for large MoE models using FP8/INT4 quantization and speculative RL for high throughput.

Enables scalable distributed LLM pretraining with torchtitan using 4D parallelism for models like Llama 3.1 and DeepSeek V3.

Comprehensive guide for writing efficient Neo4j Cypher read queries, covering modern syntax, patterns, and best practices for graph data retrieval.

Analyzes SEC filings and financial statements using EdgarTools, providing company data, XBRL financials, and multi-company comparative analysis.

Creates AGENTS.md and AI configuration files for agent setup and AI assistant configuration.

Configures pgvector for semantic search in PostgreSQL, enabling efficient text embedding similarity searches with best practices.

Optimizes LLM prompts, evaluates outputs, implements RAG systems, and structures AI workflows for enhanced model performance and reliability.

Builds scalable data pipelines, ETL/ELT systems, and data infrastructure using modern tools like Spark, Airflow, and dbt for data modeling and governance.

Builds and deploys computer vision systems for object detection and segmentation using deep learning frameworks and optimization tools.

Identifies PostgreSQL tables suitable for conversion to TimescaleDB hypertables by analyzing schema and usage patterns.

Step-by-step guidance for designing TimescaleDB hypertables with partitioning, compression, retention policies, and continuous aggregates for time-series data.