3. Data, Science & AI

Extracts study data from papers into a structured CSV table per extraction schema, enabling systematic review synthesis.

Expands and normalizes literature metadata for evidence-based surveys, producing ≥200 candidate papers with stable IDs and provenance.

Drafts non-prose visual artifacts (tables, timelines, figure specs) for surveys using evidence and BibTeX citations.

Processes raw paper data to produce a deduplicated set and a ranked core set for analysis and organization.

Provides structured configuration for OpenCode AI agents using markdown with YAML frontmatter, including mode, tools, permissions, and best practices.

Configures custom AI agents with JSON structures, tool setups, prompt patterns, and security best practices for development assistants.

Expert guidance for building, refining, and troubleshooting Claude Code subagents through agent design, system prompts, and tool configuration best practices.

Enables direct communication with OpenAI Codex by sending structured prompts and parsing JSON responses for code generation tasks.

Sends structured prompts to Gemini AI and parses JSON responses for seamless application integration.

Creates project-specific AGENTS.md configuration files from user goals with overwrite confirmation.

Creates new AI agent skills for Claude Code, Codex, and Cursor, enabling custom capabilities and slash command setup.

Manages agent memory for saving, recalling, and organizing conversational context and valuable findings.

AI-powered operating system for autonomous project loops, automated orchestration, and rule-based execution of AI workflows.

Automates research topic analysis and generates professional knowledge graphs in PDF format.

Guides AI agent development using AWS Strands Agents SDK and Amazon Bedrock AgentCore, covering architecture, deployment, observability, and multi-agent orchestration.

A Python framework for evaluating AI model quality, performance, and safety in natural language processing and generative AI workflows.

Automatically extracts and stores reusable coding patterns from Claude sessions to enhance future AI-assisted development.

Enables querying Studio deployment PostgreSQL databases for transaction debugging and analytical insights.

Formats training data in JSONL for ReinforceNow, handling entry structures, tools, rewards, and sandbox setup.

Enables on-policy distillation training using token-level KL penalty with teacher models like Qwen and Llama.

Configures ReinforceNow training runs using config.yml and train.jsonl, and converts HuggingFace datasets to required format for model training.

Generates reward functions for Reinforcement Learning (RL) training in ReinforceNow, supporting @reward decorators, precondition rewards, and LLM verification.

Generates training data from teacher models for off-policy distillation, enabling efficient knowledge transfer and SFT dataset creation.

Provides CLI for RLHF training workflows including project initialization, training runs, and model management.