Question to Learning: AI Workflow Slice

video1.0<iframe src="https://www.loom.com/embed/88d83ac7ad8f473a838273ed575e52ef" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/os/loom-media-content/loom/23bd4e8b-1408-40ec-a6f8-f078faba55e6/6ef5128e-27c0-4c71-9129-dfacb4d3dd72/sessions-thumbnails-88d83ac7ad8f473a838273ed575e52ef-9461623c160efd28.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS9iMHJxajByZGQ4MHBsNmE4IiwiYWxnIjoiUlMyNTYifQ.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.XUB-g-thF4gdEtLOVwcy8aw8J5d_e-hUumVaHgu7FgeEBSzZomjtbhN0YhIxRllLvZ3OF96Hv9KUNcoX5fJ0T95hHlNvvnfFgzeIw8QTaL7XWMuMBDDBBWg7Kis5tuGv1g74-yOKvlwv_LmsJL3D6D9vnC4PfG0xpx8CcUJcPo6WKSKZTH8QW29BuxTHIyOBdl7u5vTxY-biaUZqm9n5evrIqAB4MBGpf5oi67LNmqHW1cAosnj8R0hboGCaI8CR9tiN6mUjDK_n9OnydiUe6zRba5jpjX7mePsMZPh2xudR5SybfSl8G9pSzA_0Zg7urMiylh7_cBv3ATpoN-PATg&Expires=1792931081&Signature=oU9cTS0ys20POYyXNFCnO4mHSsUaen1FCH3PDF9rCfrnvyj2GerFr8vyzmaFk%7EzQmfP17QQymwtBr6rQ7g5k5RkjX1XqoQ-sLuzTnkSBEd-1kfO9RGyNN7uz%7EPe1e6FAUq4m3Kvwr07NYiql3-16wh37eWskLI1KQ4QPqM5fCrtpiRIJWSGSgwHBF%7EdCJecjaSUfR%7E6hOGn0niwl2eGBEk2XO5EHXOArtHnMX87euTyFWpv4pe6rYkUib10GhnVRvwGIVBRShWkk0vtakw67vefodcCZSzQ0X3GRrrhi9MKN8VTxK8Jrv2lVyX8UlaXKOh0YwnDxY85HjeZ9bAOQXw__&Key-Pair-Id=KQOSYIR44AIC0245.76Question to Learning: AI Workflow SliceThis Loom presents Diplab, a question-to-learning workflow designed to show how an AI system handles missing parts of an incomplete question. It uses five stages, Ask, Clarify, Define, Test, and Learn, with parsing by an LLM and Grok added first, falling back to Gemini second, and it caps follow-up questions at three while showing defaults and missing field labeling. The speaker evaluates the approach on 91 dipped signals over 2 years of simulated NIFTY-like data, reporting average 5-day forward returns net of cost of -0.86 for the high wall and -0.28 for the low wall. They emphasize that numbers displayed come from deterministic code with server-side clamping, and that the assignment tests the research workflow rather than training strategy.