platform/aiml/AGENTS.md

Repository Guidelines

Project Structure & Module Organization

• etl/: Core pipelines (ingestion, transformation, delivery). Key subfolders: corpus-pipeline/, xet-upload/, bleeding-edge/, config/, team/.
• mlops/: MLOps workflows, deployment and model lifecycle assets.
• training/: Training utilities and scripts.
• experiments/: Prototypes and one-off investigations.
• models/: Model artifacts/weights (large files; avoid committing new binaries).
• 07_documentation/: Internal docs and process notes.

Build, Test, and Development Commands

• Environment: Use Python 3.10+ in a virtualenv; copy .env from example or set required keys.
• Install deps (per module):
  • pip install -r etl/corpus-pipeline/requirements-scrub.txt
• Run pipelines:
  • python etl/master_pipeline.py
  • python etl/corpus-pipeline/etl_pipeline.py
• Run tests (pytest):
  • python -m pytest -q etl/
  • Example: python -m pytest -q etl/corpus-pipeline/test_full_integration.py

Coding Style & Naming Conventions

• Python (PEP 8): 4‑space indent, snake_case for files/functions, PascalCase for classes, UPPER_SNAKE_CASE for constants.
• Type hints for new/modified functions; include docstrings for public modules.
• Logging over print; prefer structured logs where feasible.
• Small modules; place shared helpers in etl/.../utils when appropriate.

Testing Guidelines

• Framework: pytest with test_*.py naming.
• Scope: Unit tests for transforms and IO boundaries; integration tests for end‑to‑end paths (e.g., etl/corpus-pipeline/test_full_integration.py).
• Data: Use minimal fixtures; do not read/write under etl/corpus-data/ in unit tests.
• Target: Aim for meaningful coverage on critical paths; add regression tests for fixes.

Commit & Pull Request Guidelines

• Commits: Use Conventional Commits, e.g., feat(etl): add scrub step, fix(corpus-pipeline): handle empty rows.
• PRs: Include purpose, scope, and risks; link issues; add before/after notes (logs, sample outputs). Checklists: tests passing, docs updated, no secrets.

Security & Configuration Tips

• Secrets: Load via .env and/or etl/config/etl_config.yaml; never commit credentials.
• Data: Large artifacts live under etl/corpus-data/ and models/; avoid adding bulky files to PRs.
• Validation: Sanitize external inputs; respect robots.txt and rate limits in crawlers.

Architecture Overview

graph TD
  A[Ingestion\n(etl/ingestion + crawlers)] --> B[Transformation\n(clean, enrich, dedupe)]
  B --> C[Storage & Delivery\n(JSONL/Parquet, cloud loaders)]
  C --> D[MLOps/Training\n(mlops/, training/)]
  D -- feedback --> B

Key flows: etl/corpus-pipeline/* orchestrates raw → processed; delivery targets live under etl/xet-upload/ and cloud sinks.

CI & Badges

Add a simple test workflow at .github/workflows/tests.yml that installs deps and runs pytest against etl/.

Badge (once the workflow exists): ![Tests](https://github.com/OWNER/REPO/actions/workflows/tests.yml/badge.svg)

Minimal job example:

name: Tests
on: [push, pull_request]
jobs:
  pytest:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with: { python-version: '3.10' }
      - run: pip install -r etl/corpus-pipeline/requirements-scrub.txt pytest
      - run: python -m pytest -q etl/