3. Data, Science & AI

Manages Supabase databases, authentication systems, and backend services for project data infrastructure.

Manages PostgreSQL database schema changes securely, including diagnosis, Alembic revision generation, and integrity validation.

Enables interaction with DeepSeek V3/R1 language models for high-performance reasoning and coding, compatible with OpenAI API standards.

Enables interaction with Kimi (Moonshot AI), a large language model optimized for extended context handling and coherent long-form dialogue.

Accesses Amap (Gaode) API for POI search, weather data, and IP geolocation services in China.

Provides a Python SDK for building production-ready AI agents with autonomous workflows, custom tools, and context management.

Retrieves real-time weather forecasts for China using the QWeather API (HeFeng).

Creates end-to-end machine learning pipelines for data preprocessing, model training, and performance evaluation.

Creates MCP servers enabling LLMs to interact with external services through well-designed tools, covering Python (FastMCP) and TypeScript (MCP SDK) implementation.

Provides performance optimization patterns for PostgreSQL using psycopg, specifically for batch data operations and query efficiency.

Develops ML models for RetentionAI, including CLV prediction (BG/NBD), uplift modeling, and feature engineering pipelines.

Guides creation of MCP servers for LLMs to interact with external APIs, supporting Python (FastMCP) and Node/TypeScript (MCP SDK).

Enables Ollama LLM integration for Chuukese translation and text generation with custom modelfiles and Helsinki-NLP hybrid workflows.

Advanced web scraping tool for bypassing bot protections and extracting structured data from complex web pages using AI-powered extractors.

Provides advanced techniques for transforming raw data into optimized features to improve machine learning model performance and accuracy.

Processes Bible EPUBs (NWT) to extract structured verse data for parallel text alignment between Chuukese and English, enabling corpus building and AI training data generation.

Orchestrates autonomous AI agent loops for executing multi-step tasks including coding and content creation without human intervention.

VidLens extracts structured video artifacts including metadata, frame snapshots, and scene cuts using ffmpeg and Bun for video analysis.

Provides database schema design, ETL pipeline development, data quality validation, and query optimization for RetentionAI's data infrastructure.

Generates AI images via Gemini CLI with secure Keychain API key management for safe, efficient image creation.

Enables meaning-based information retrieval from GTM knowledge base, moving beyond keyword matching to contextual relevance.

Creates and adjusts AI opponent behaviors for games or simulations, emphasizing strategic decision-making and adaptive responses.

Creates and analyzes evaluations for AI agents and LLM applications, with recommended EZVals framework for testing and debugging.

Builds evaluation frameworks to test AI agent performance, validate context engineering, and measure iterative improvements.