| # Elizabeth AIML Platform — Nova R&D (Soul + Mask + Fast‑Weights) | |
| This repo contains the Elizabeth AIML platform codebase and the Nova R&D blueprint. The goal is a single lifelong agent with identity anchored in weights (Soul), safe real‑time plasticity (Mask ≤5%), and immediate stickiness via Fast‑Weights — with rigorous receipts, eval gates, and rollback. | |
| Key locations: | |
| - `projects/elizabeth/blueprint/`: R&D blueprint, ADRs, experiments, metrics, receipts. | |
| - `mlops/`: gateway, tools, receipts, sync scripts. | |
| - `etl/`: pipelines and data utilities. | |
| - `models/`: model artifacts (do not commit large binaries to GitHub). Use Hugging Face for artifacts. | |
| Sync policy: | |
| - Code → GitHub `adaptnova/e-zeropoint` (private). Branches: `main`, `develop`. | |
| - Artifacts → Hugging Face `LevelUp2x/e-zeropoint` (private). LFS for weights; publish via `mlops/sync/publish_hf.sh`. | |
| Auth & secrets: | |
| - GitHub: authenticated via `gh` CLI (see `gh auth status`). | |
| - Hugging Face: set `HUGGINGFACE_HUB_TOKEN` in `/data/adaptai/secrets/dataops/.env`. | |
| Receipts & Ops: | |
| - Per‑turn receipts under `projects/elizabeth/blueprint/13_receipts/` and Slack summaries if configured. | |
| - See `mlops/receipts/collect_receipt.py` and `mlops/slack/post_update.py`. | |
| Contribution: | |
| - Python 3.10+, type hints on new functions, logging over print. Tests under `etl/` with `pytest`. | |