3. Data, Science & AI

Analyzes session performance to iteratively refine agent prompts, enabling continuous self-improvement and optimization of AI agent behavior.

Logs LLM context window usage to JSON Lines file upon detecting 'log context' in prompts, aiding AI interaction monitoring.

Captures structured agent trajectories (thoughts, tools, plans) from headless CLI agents for AI evaluation via JSONL output.

Secure PostgreSQL CLI enabling AI agents to query databases, list tables, and inspect schemas without credential handling.

Manages acquisition session data including preprocessing, migration between projects, and deletion with user confirmation via the sl-manage MCP server.

Expert guidance on database design, query optimization, and data integrity for both SQL and NoSQL systems.

Converts ML models (ONNX, TFLite, PB, PyTorch) to QNN context binaries for Qualcomm hardware deployment.

Guides setup and operation of local LLMs using Harbor, covering Ollama, Open WebUI, llama.cpp, and vLLM for model deployment.

Enables high-performance scientific charting and data analysis using SciChart Engine with WebGL rendering.

Guides users in creating specialized AI agents (subagents) through configuration guidance and architectural best practices for agent-based systems.

Manages LLM traces, prompts, datasets, and evaluation scores via Langfuse API for AI development and model analysis.

Adds observability to LLM applications using Langfuse for tracing and auditing LLM calls.

Automates migration of hardcoded prompts to Langfuse for version-controlled prompt management in LLM applications.

Enables creation and configuration of custom subagents for Claude Code, supporting specialized AI assistants and task delegation via file-based agents or Task tool invocation.

Analyzes CSV/Excel datasets for descriptive statistics, data distribution, feature exploration, and anomaly detection to support data-driven insights.

Manages data integrity for agent chat data using IndexedDB v3 with Optimistic Updates and Eventual Consistency patterns.

Manages AGENTS.md documentation to optimize AI coding agent performance and context window usage in LLM-based development workflows.

Manages HuggingFace datasets for Security Verifiers, including pushing, metadata, gated access, and repository setup for E1/E2 datasets.

Builds and manages datasets for security verification, including E1/E2 datasets, test fixtures, and data validation for network logs and configuration environments.

Enables rapid tool integration for AI agents using Composio's routing system to enhance agent functionality.

Provides context packing and anti-duplication guardrails to optimize LLM context handling and prevent redundant data in AI workflows.

Coordinates AI agents using Swarm mail messaging and file locking to prevent conflicts during collaborative tasks.

Enables SQL-based analysis of Claude Code conversation history, token usage, and agent activity from session logs.

Converts Product Requirements Documents (PRDs) into Ralph's required JSON format for autonomous agent input.