3. Data, Science & AI

Provides headless CLI access to Google's Gemini AI for code generation, analysis, and multi-modal reasoning tasks beyond Claude's native capabilities.

Optimizes prompts for Claude AI to enhance clarity, effectiveness, and desired outputs in AI interactions.

Simplifies crypto data filtering by eliminating redundant checks and prioritizing essential data quality, spread, and trading status validations.

Saves backtest results to SQLite for tracking history, comparing model performance, and querying top results.

Enables semantic search and RAG implementation using Qdrant vector database with embedding generation, HNSW indexing, and OpenAI integration.

Adjusts deconvolution output scaling to preserve channel intensity ratios and prevent saturation issues for accurate quantitative image analysis.

Fixes excessive HOLD bias in RL trading models by adjusting reward penalties for low trade rates and slippage.

Performs statistical analysis to select trading symbols compatible with mean-reversion/momentum strategies using GARCH risk modeling and Markov regime detection.

Fixes common PyTorch bugs in neural network and reinforcement learning code, including percentile calculations and LayerNorm for Conv1d.

Integrates OpenAI GPT models for conversational AI, supporting chat completions, function calling, and agent persona development with cost optimization.

Parses multi-format channel names from KINTSUGI CHANNELNAMES.txt files, enabling structured data extraction and standardization.

Provides a router for domain-specific subagents, enabling creation, auditing, and refinement for isolation and parallelism in AI agent systems.

Enables connecting Jupyter notebooks in data pipelines via checkpoint files for output propagation and multi-notebook workflow management.

Normalizes trading pattern filters and enables multi-gate order execution for automated trading strategies.

Visualizes sparse single-cell gene expression data using specialized plots to highlight zero-inflated patterns and expression distributions.

Ensures Alpaca API is prioritized for financial data, avoiding fallback to yfinance under specific quality conditions.

Permanently caches yfinance sector data to bypass rate limits, triggered by 401 errors, slow updates, or selective refresh needs.

Optimizes machine learning model training with learning rate warmup, adjusted validation intervals, and reward weighting to address instability and conservatism.

Converts backtest visualization axes from bar indices to actual datetime for improved time context in financial data analysis.

Expert assistance with Microsoft SQL Server database development, including T-SQL queries, schema design, performance tuning, and Node.js integration.

Provides a consistent drawdown control pattern for trading systems, applicable in backtesting, live trading, and model training to manage risk.

Automatically optimizes PPO training configuration for A100/H100 GPUs to resolve slow training and low GPU utilization.

Executes post-training validation steps including gating, backtesting, and walk-forward validation to determine model deployment readiness after GPU training.

Embeds version metadata in model checkpoints and validates compatibility during loading to prevent failures from version divergence or interpretation issues.