3. Data, Science & AI

Provides detailed rules for SQL schema design, 3NF normalization, Row Level Security (RLS), and triggers in Supabase.

Indexes and searches project documentation using semantic similarity to find relevant content and link to code entities.

Creates reactive notebooks for data analysis and interactive visualizations using the Marimo framework.

Starts fresh sessions when context becomes too complex while preserving continuity through episodic memory storage.

Generates images from text prompts and edits existing images using Google's Gemini API for AI-driven visual content creation.

Optimizes 3D Gaussian Splat scenes for real-time rendering on Apple platforms using LOD, pruning, and compression strategies.

Fine-tunes LLMs via Unsloth with SFT, GRPO, FP8, vision models, and GGUF export capabilities.

Optimizes CoreML models for iOS/macOS deployment through quantization, pruning, and Neural Engine targeting to reduce size and latency.

Enables vector search, embedding storage, and similarity calculations in PostgreSQL using pgvector and HNSW indexing for AI/ML applications.

Enables building multi-agent workflows with LangGraph, supporting supervisor/swarm patterns and cost-optimized orchestration across multiple LLM providers.

Expert guidance for configuring Gemini CLI, crafting advanced prompts, and creating context files for AI agent interactions.

Provides fast LLM inference via Groq API for chat, vision, audio STT/TTS, and tool calling in real-time applications.

Enables production voice AI agents with sub-500ms latency using Groq LLM, Deepgram STT, and Cartesia TTS for phone bots and IVR systems.

Provides expert guidance for building AI agents using DeepAgents framework with seamless LangChain/LangGraph tool integration.

Guides construction of stateful, multi-actor AI agents using LangGraph's graph structures, nodes, edges, and state management.

Manages map layers, GeoJSON sources, and styling for geospatial data visualization using MapLibre. Handles layer interactions and dynamic composition.

Manages candidate scoring, confidence calculation, and decision logic using AI/ML techniques such as softmax, temperature tuning, and entropy thresholds.

Tests game mechanics by querying game data via SQL, utilizing game_* tools for gameplay flow validation.

Manages trait extraction, place enrichment, and system learning from gameplay data, including LLM prompt configuration and async queue processing.

Provides REST API access to extract and transform operational data from Knack objects and views for data integration workflows.

Enables geographic query processing, spatial filtering, and geometry operations using PostGIS functions and indexes in PostgreSQL.

Analyzes trading strategy backtest results to identify weaknesses and offers data-driven suggestions for optimization and improved performance.

Performs data validation, deduplication, normalization, and integrity checks to ensure accuracy for HTI compliance and performance dashboards.

On-device specialist for validating audio machine learning models in Modcaster's AI-driven audio processing pipeline.