3. Data, Science & AI

Sets up autonomous Claude code loops for continuous task execution until completion using trigger phrases.

Analyzes codebases to reduce token usage for LLM efficiency, reporting waste patterns and suggesting optimizations to lower costs and respect context limits.

Converts JSON data to compact TOON format to minimize token usage, optimizing data for LLM processing and reducing AI workflow costs.

Computes and visualizes stable manifolds in dynamical systems, representing points converging to equilibrium.

Guides memory-efficient API usage patterns for LLM evaluations and batch processing, avoiding common anti-patterns.

Automates ML experiment logging with Hydra configuration and Inspect AI model evaluation.

Generates reproducibility reports for LLM/ML experiments by auto-extracting from code, configs, and Hydra, identifying missing documentation gaps.

Runs AI/ML experiments with output isolation to prevent context pollution, ensuring clean and reproducible results.

Provides a mock prediction environment for 4-layer thinking tests, utilizing a subagent to simulate task execution.

Fetches financial news articles for a specific stock ticker within a specified date range, including optional month window filtering.

Analyzes news events to identify drivers of stock price movements, providing confidence scores and expected returns.

Simulates attribution for 4-layer thinking tests by delegating to a Task subagent in an AI reasoning workflow.

Validates token capture during LLM reasoning to ensure accurate internal thinking process tracking.

Executes an MCP query in the bottom layer of a three-layer test framework for AI model interaction.

Orchestrates batch analysis of SEC 8-K filings with predictive modeling and attribution accuracy tracking for earnings insights.

Fetches Benzinga financial news with significant daily returns from a Neo4j graph database for data analysis and market insights.

Mock orchestrator for testing 4-layer reasoning in AI models through coordinated prediction and attribution skill execution.

Starts coordinated multi-agent tasks using hive-mind coordination for complex research and implementation.

Manages session memory for pseudo-code prompting, preserving context, learned patterns, and user preferences across interactions.

Converts verbose natural language queries into structured, code-style pseudo-code for optimized AI prompt engineering and direct model communication.

Automates a data pipeline that transforms, validates, and optimizes data in a single, chained process.

Applies prompt engineering best practices to CLAUDE.md, SKILL.md, and slash commands for optimized AI interactions.

Monitors status of AI subagents, tracking token usage and context limits for optimal language model performance.

Reads and searches digital books (PDF, EPUB, MOBI, TXT) to extract information, find quotes, and answer book-related questions.