AI Engineering

Design Patterns helps agents choose implementation patterns and abstraction boundaries across Rust, TypeScript/React, and Django/Python. It covers when to use, when to avoid, and trade-offs for each option.

Feature Development guides a new feature through codebase understanding, clarifying questions, architecture design, implementation, and quality review. Use it when you need a methodical way to add functionality instead of jumping straight into code.

HF Mem estimates the VRAM or memory needed to load Hugging Face Safetensors or GGUF model weights for inference. Use it to check whether a model fits on a GPU or instance, and optionally include KV cache estimates.

Hugging Face Community Evals runs inspect-ai and lighteval evaluations for Hugging Face Hub models on local hardware. Use it to choose an inference backend, smoke-test tasks, and run GPU evals with vLLM, Transformers, or accelerate.

HuggingFace Best Model Finder finds recommended Hugging Face models for a task by checking official benchmark leaderboards and model size. It helps when you need to compare candidates against device constraints or choose a top model to try.

CASS Memory System provides procedural memory for AI coding agents with cross-session and cross-agent context retrieval. Use it before non-trivial tasks to surface relevant rules, anti-patterns, and past session history.

Deepdive turns a research question into a documented multi-step investigation with sources, hypotheses, and cited Q&A reporting. Use it for desk research, hypothesis validation, and understanding how something works.

Local vision CLIs (glance, ground, detect, trace, crop) give text-only agents the ability to describe, OCR, locate elements, extract shapes, and crop images. Also includes scripts for long screenshot OCR, HTML-to-image, and more. Use for any task involving an image — from reading text to GUI automation.

Choose Covariance Estimator picks a covariance estimator for a 2-D dataset using sample features like dimension, conditioning, and tail behavior. Use it when you need a shortlist of estimators or a safe default for ill-conditioned or heavy-tailed data.

Fin Guru Strategize turns quantitative analysis into actionable portfolio recommendations. Use it to combine margin, dividend, and cash-flow tactics into a risk-checked wealth-building plan.

Finance Guru Core Context loads the system configuration, user profile, and related files at session start. Use it to ensure full financial context is available before analysis, planning, or portfolio operations.

Learner Profile builds progressive learner and client profiles for Finance Guru onboarding and teaching. It captures skill level, goals, preferences, and ADHD-friendly accommodations to personalize sessions.