# AI_Brain

Seeding suite for Hive's 500× expansion. Inserts atomically hashed sentences into Postgres `atomic_facts` with ON CONFLICT DO NOTHING.

## Features
- 12-factor env config; dry-run support via `DRY_RUN=1`.
- File-based seeding (`SEED_FILE[_CATEGORY]`) or programmatic generators (math, agent, language, prophecy, marketing).
- Online marketing intelligence endpoint that ranks curated campaigns against prompts.
- Statistical marketing intelligence endpoint delivering benchmark-weighted ROI and CTR projections.
- Multi-process marketing ANI surface pairing retrieval-augmented guidance (RAG), an Experiment Growth Grid (EGG) planner, and optional Google Ads enrichment.
- Atomic SHA-256 hashing and idempotent inserts.
- Simple launcher: `daemon/seed_all.sh --env file --dry-run --only math,agent,marketing,...`.

## Quick Start
1. Python deps (for live DB):
   - `python3 -m pip install --user -r requirements.txt`
2. Create schema:
   - `psql -h "$PGHOST" -U "$PGUSER" -d "$PGDATABASE" -f infra/seed.sql`
3. Dry-run with examples:
   - `daemon/seed_all.sh --env .env.example --dry-run`
4. Live (set your own `.env`):
   - `daemon/seed_all.sh --env .env`

## Env Vars
- PGHOST, PGDATABASE, PGUSER, PGPORT
- DRY_RUN (1 to simulate)
- SEED_SOURCE (default `Hive500Expansion`)
- SEED_FILE, SEED_FILE_MATH_MODEL, SEED_FILE_AGENT, SEED_FILE_LANGUAGE, SEED_FILE_PROPHECY, SEED_FILE_MARKETING
- GOOGLE_ADS_DEVELOPER_TOKEN, GOOGLE_ADS_CLIENT_ID, GOOGLE_ADS_CLIENT_SECRET, GOOGLE_ADS_REFRESH_TOKEN, GOOGLE_ADS_CUSTOMER_ID, GOOGLE_ADS_LOGIN_CUSTOMER_ID (optional; enables live Google Ads metrics in the marketing ANI)

See `.env.example` for a template.

## Make Targets
- `make enumerations` – generate 5×500 lists under `daemon/enumerations/`
- `make seed-dry` – dry-run using `.env.example`
- `make seed-live` – live run using `.env` (ensure DB is reachable)

## Testing
- `python -m pip install -r requirements-dev.txt`
- `pytest`

## Intelligence API

The marketing ANI runs within the FastAPI service and now includes a statistical surface.

- Run locally: `uvicorn services.seeder_api.app:app --reload`
- Baseline ranking endpoint: `curl -X POST http://localhost:8000/intelligence/marketing -H 'Content-Type: application/json' -d '{"prompt": "accelerate demos for B2B SaaS founders"}'`
- Statistical endpoint with ROI/CTR projections: `curl -X POST http://localhost:8000/intelligence/marketing/statistical -H 'Content-Type: application/json' -d '{"prompt": "Optimize ROI for AI founders"}'`
- Statistical responses include campaign-level ROI, CTR, confidence estimates, and a portfolio rollup summarizing the blended strategy.
- Diagnostics endpoint for prompt intelligence: `curl -X POST http://localhost:8000/intelligence/marketing/diagnostics -H 'Content-Type: application/json' -d '{"prompt": "Pipeline is stalled for B2B founders"}'`
- Diagnostic responses label funnel stage, urgency, ICP focus, recommended KPIs, opportunity plays, and risk signals to accelerate decisioning.
- Process orchestration endpoint for RAG + EGG + Google Ads: `curl -X POST http://localhost:8000/intelligence/marketing/processes -H 'Content-Type: application/json' -d '{"prompt": "Need urgent conversion plays for enterprise SaaS"}'`
- Process responses include a `rag` block with prioritized campaigns, an `egg` array of experiment playbooks, and a `google_ads` collection containing either live Google Ads performance (when credentials are configured) or benchmark-driven simulations.

## Continuous Integration
This repository is wired up with a GitHub Actions workflow (`CI`) that runs on
every push and pull request. The workflow installs the project dependencies,
executes the dry-run seeding smoke test (`make seed-dry`), runs the automated
pytest suite, and compiles the Python sources to validate imports. This provides
quick feedback that seeding pipelines and services stay healthy before changes
are merged.