Design and optimize AI agent action spaces, tool definitions, and observation formatting.

Systematic self-debugging workflow for agents experiencing repeated failures, loops, or context drift.

Act as a neutral arbiter to resolve merge conflicts between two agent branches.

Coordinate C-suite agent teams with structured invocation syntax, loop prevention, and validation rules.

Design robust AI agent tools with expert guidance on schemas, descriptions, and error handling.

Manage AI-driven development with eval-first execution, task decomposition, and model routing.

Best practices and guidelines for hyperparameter searches and AutoML tooling with Optuna, Ray Tune, and PyCaret.

Reference guide and workflow for building LLM-powered applications with the Claude API and Anthropic SDK.

Analyze a codebase and recommend tailored Claude Code automations across hooks, subagents, skills, and MCP servers.

Audit context window consumption across agents, skills, MCP servers, and rules to reclaim space.

Optimize LLM API costs with intelligent model routing, immutable budget tracking, and prompt caching.

Run a board memo or strategy brief through multiple model providers to reconcile divergences and achieve consensus.

Configure deep agent architectures, built-in middleware, backends, and modular skills for LangChain.

Configure short-term, long-term, and hybrid storage backends for deep agent file operations.

Orchestrate deep agents with subagent task delegation, todo list planning, and human-in-the-loop approvals.

Guide LLM fine-tuning, dataset preparation, PEFT adapter configuration, and model evaluation workflows.

Implement on-device LLM text generation, guided structured output, and custom tool calling in iOS.

Analyze markdown files and suggest optimizations to reduce token consumption while maintaining clarity.

Follow a phased methodology to add machine learning algorithms to existing non-ML codebases.

Build production ML systems with robust data contracts, reproducible training, and operational monitoring.

Conduct a founder interview to generate the company context file required by C-suite advisors.

Apply idiomatic PyTorch patterns for reproducible training loops, clean architectures, and data pipelines.

Run structured recursive decision loops with append-only ledgers and clear promotion rules.

Optimize parsing costs by combining regex for predictable text patterns with LLMs for edge cases.