New Mexico Concierge (Mockup)

A production-minded FastAPI + Jinja2 app that renders a New Mexico Concierge chat page and serves a RAG-backed chatbot grounded in New Mexico Magazine archives and New Mexico True Certified listings.

Run locally

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
uvicorn app.main:app --reload

Visit http://localhost:8000.

Ingest content

1. Start the app.
2. Set ADMIN_TOKEN in your .env and export it in your shell.
3. Trigger ingestion:

export ADMIN_TOKEN=change-me
./scripts/ingest_seed.sh

Optional: pass a base URL to the script (e.g. ./scripts/ingest_seed.sh http://localhost:8000).

Allowlist/denylist controls live in config/urls_allowlist.yaml and config/urls_denylist.yaml. The allowlist is prefilled for New Mexico Magazine (all pages) and New Mexico True Certified sections. Crawling respects robots.txt (both sites allow crawling with a 2-second delay).

Set CRAWL_MAX_PAGES in .env to control crawl scope (default 2000).

Ingest status (admin)

Visit http://localhost:8000/admin/ingest?token=ADMIN_TOKEN or send Authorization: Bearer ADMIN_TOKEN to view a simple ingestion status page.

GeoLite2 City database

Set GEOLITE2_CITY_DB_PATH to the path of the GeoLite2 City .mmdb file. If the file is missing or the setting is empty, the app stores NULL for location fields and continues processing requests.

Deployment notes

• Container-friendly: run with uvicorn app.main:app --host 0.0.0.0 --port 8000.
• Persist app.db and data/faiss.index between deploys.
• Provide OPENAI_API_KEY, OPENAI_MODEL, OPENAI_EMBED_MODEL, ADMIN_TOKEN, and optional ALLOWED_ORIGINS.

Brand assets (optional)

Place official logos or images in app/static/brand/. The scripts/fetch_brand_assets.py helper can download assets once URLs are provided.

Privacy

Analytics store timestamps, message text, model used, retrieval count, latency, user agent, approximate token counts, and approximate city/region/country. IP addresses are never stored or hashed.

Cost notes

The default model is gpt-4.1-mini for quality. You can switch to a cheaper model via OPENAI_MODEL for load testing.