3. Data, Science & AI

Advises on token budgeting for AI tasks involving large files, multi-document analysis, and complex workflows to prevent token limit exceedance.

Builds multi-tab Excel workbooks with complex calculations, pivot tables, and visualizations for startup financial models and board reporting.

Analyzes skill usage patterns to track token consumption, identify heavy/lightweight skills, and suggest optimization opportunities for efficient AI assistant operation.

Seeds a mahresources database with semantic content across all entity types, generating over 50 items per type to enable pagination testing.

Provides natural language understanding, intent classification, and context management for agent-based conversational systems with history analysis.

Activates traditional knowledge structures through semantic activation using K-line patterns for AI knowledge representation.

Builds, lints, ingests, and composes Drescher-style schemas to ensure consistent data modeling and validation.

Orchestrates parallel AI agent execution with budget controls and quality gates via Sibyl's MetaOrchestrator.

Graph-RAG knowledge system for semantic search, task tracking, and project audits with persistent memory across sessions using FalkorDB.

Configures, develops, and deploys AI agents and skills within the Claude Code ecosystem, supporting CLI tools, MCP servers, and CI/CD integration.

Integrates Google AI tools for multimodal processing including transcription, image analysis, and document queries using Gemini API and NotebookLM.

Evaluates AI-generated code quality using pass@k metrics, supporting capability and regression testing for code generation models.

Optimizes LLM context windows through compression, tool hooks, and mode injection for efficient agent workflows.

Assesses codebases for compatibility with autonomous AI agents, ensuring adherence to best practices for unattended execution and agent-friendly design.

Diagnoses AI feature issues including hallucinations and inconsistency through 4D root cause analysis, targeting model behavior and data flow.

Generates first 20 AI evaluation test cases in spreadsheet using PM-Friendly Evals approach, avoiding overengineering.

Pre-launch health check for AI features that grades model selection, data quality, cost, monitoring, failure UX, and optimization to prevent deployment of broken models.

Enables creation and management of AI agents for complex task automation using natural language interfaces.

Streamlines AI agent application development with code generation, model comparison, tracing, and evaluation setup for efficient workflow creation.

Executes SQL queries on Doris database clusters via API, supporting batch execution from files or multiple statements, and returns results or status.

Reads user profile data and calculates content relevance to personalize outputs based on user interests.

Performs data cleaning, statistical analysis, visualization, and insight generation from datasets.

Provides a Python SDK for AI-driven data processing of dataframes via natural language instructions for transformation, deduplication, and merging.

Generates AI-powered images via Google Gemini with customizable parameters for creative projects.