YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
NotGPT Phase 1 β MVP Design & Build Handoff
Design review, not build. Commissioned 2026-11-17 by the platform handoff card. These documents re-derive the target MVP from the as-built AWS codebase (bucket david-thrower/NotGPT-Phase-1-temp) and current Hugging Face / model-provider pricing (read 2026-11-17). Nothing here has been built; the build must not be fired until the cost model (Β§8 of the design doc) is re-verified against current prices.
Documents
- design-doc.md β the refactored MVP design: as-built architecture with file attribution, the discrepancy list (upsampling was k=1, 60/40 subsetting was a stub, the reasoning column was dropped, personal-slm served a base model), the AWS β Hugging Face mapping table, the corrected personalization pipeline as an HF Jobs DAG, the deployed agent with configurable frontier tiers, the tertiary-detail-recall eval, and the full cost model in both regimes (Scenario A: Nemotron 3 4B student; Scenario B: HelixLM 300β500M student) with sensitivity and break-evens.
- build-handoff-card.md β the skill for a fresh agent: mission, non-negotiables, the verification gate, 10 build phases with acceptance gates, the as-built β must-fix list, per-phase cost guardrails, acceptance criteria, and sign-off items.
The numbers in one breath
| Scenario A (4B student) | Scenario B (HelixLM 300M student) | |
|---|---|---|
| Build (agent tokens + compute) | ~$60β450 (driver-dependent) | same |
| Onboarding / user (250 MB corpus) | $75β96 β over the <$50 target | $25β46 β |
| Monthly / user online | ~$3β84 (frontier tier) | ~$3β84 (frontier tier) |
The product case: scenario B with an API-hosted teacher hits the <$50 onboarding target everywhere; scenario A does not except at small corpora.
Source material
- Old codebase:
hf://buckets/david-thrower/NotGPT-Phase-1-temp/NotGPT-Phase-1-temp-latest-all-2-to-hf/NotGPT-Phase-1-temp-latest-all/(public as of 2026-11-17) - HelixLM base model:
david-thrower/HelixLM-53.9M-20261116-br72-d512-h8-nl3-s2048-5.0BT-wsd
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support