3. Data, Science & AI

Optimizes LLM cost and latency through caching, model selection, batching, and prompt engineering, providing cost breakdowns and recommendations.

Diagnoses LLM output failures including hallucinations, format errors, and reasoning flaws with root cause analysis and prompt fixes.

Designs reliable ETL and data synchronization jobs with incremental updates, idempotency, and error handling.

Automates detection and resolution of database integrity issues including orphaned records and broken foreign keys.

Sets up vector databases for semantic search and embedding storage, supporting Pinecone, Chroma, pgvector, and Qdrant.

Enables advanced time interval handling, partition pruning, and time formatting for efficient data filtering and query optimization.

Builds LLM agents using YAML/Markdown configuration with tools for knowledge bases, web search, and image generation in the TD AI workflow.

Provides professional Plotly visualization best practices including chart specs, color palettes, and JSON structure for executive-ready data visualizations.

Provides TD-specific Trino SQL functions for time-based filtering, partition pruning, and TD query patterns.

Validates CDP journey YAML configurations against tdx schema to ensure correctness before deployment to Treasure Data.

Converts Trino SQL queries to Hive syntax, addressing memory errors and syntax differences like td_time_string to TD_TIME_FORMAT.

Provides command-line utilities for Treasure Data database management, table exploration, query execution, and authentication setup.

Crafts optimized system prompts for AI agents, including role definition, constraints, output formatting, and tool instructions to enhance agent performance.

Expert assistance for using pytd Python SDK to query, import, and analyze data in Treasure Data, supporting Presto/Hive and pandas integration.

Configures dbt for Teradata Trino data transformation, including connection setup, model overrides, and workflow deployment.

Enables Hive SQL processing with TD-specific functions and Hive-only features like LATERAL VIEW and MAPJOIN for data analysis tasks.

Validates Treasure Data CDP segment YAML configurations against the API specification to ensure correctness and prevent deployment errors.

Manages CDP audience segments via YAML rule configurations, supporting filtering, validation, and export to marketing platforms.

Generates connector configuration for Segment and Journey activations by using the `tdx connection schema` command to discover available fields.

Manages Treasure Data workflows using Digdag syntax, session variables, and workflow directives for data pipeline orchestration.

Manages CDP parent segments via CLI commands, including validation, preview, and scheduling for data segmentation workflows.

Optimizes Trino queries for performance using CTAS, UDP bucketing, and function comparisons to accelerate data processing.

Consolidates channel transcripts into a single sorted file (newest first) for LLM context or bulk analysis, with an 800K token limit.

Extracts YouTube channel video metadata (title, description, etc.) and saves to CSV for tracking and management purposes.