A local research terminal wiring market data, storage and backtesting behind seven MCP servers; its walk-forward pipeline validates strategy grids over decades of data with in-sample selection and out-of-sample stitching.
Episode specs become finished videos with zero manual editing: TTS narration, HTML scene rendering, headless-Chromium capture, and burned-in captions from one JSON source of truth.
Converts real HKDSE maths papers into structured item banks, then generates topic- and difficulty-profiled mock papers in LaTeX — with distractor-to-error mapping that turns wrong MC answers into revision signals.
This site: a capability-tiered hero with a rigged 3D robot tracking your cursor, Payload-CMS-editable content, and a Lighthouse mobile 0.99 budget held through every feature.
Modular CLI computing readability and lexical-diversity metrics for English and Chinese academic text, with a reusable prompt library and automated quality gates (300+ tests).
A 10-profile Hermes AI-agent environment: routes requests across multiple LLM providers with cost-based fallbacks, orchestrates parallel research subagents, and packages reusable skills as tested Python CLI tools.