nl_config.py

"""VynFi NL config generation — describe a dataset, get a validated config.

Two sibling endpoints (both Scale+):

- ``client.configs.from_description(text)``  — free-text description →
  validated PortalGenerationConfig.
- ``client.configs.from_company(uid=..., name=...)``  — Swiss VynCo
  company profile → config tailored to the real entity's profile.

This example exercises both, prints the generated config, and (optionally)
submits it for a tiny dry-run generation so you can see the round-trip.
"""

from __future__ import annotations

import json
import os

import vynfi

client = vynfi.VynFi(api_key=os.environ["VYNFI_API_KEY"])

# ── 1. From free-text description ────────────────────────────────────────────

description = (
    "Swiss manufacturing midsize firm, 4 quarterly periods in 2025, "
    "IFRS, CHF, 3 legal entities, fraud-light baseline with a "
    "vendor-kickback scheme, AML-free. Include audit workpapers."
)

print("=== configs.from_description() ===")
nl = client.configs.from_description(description)
print(f"confidence: {nl.confidence}")
print(f"notes:      {nl.notes}")
cfg_preview = {k: v for k, v in (nl.config or {}).items() if not k.startswith("_")}
print("generated config (top-level keys):")
for key in sorted(cfg_preview):
    print(f"  - {key}: {json.dumps(cfg_preview[key], default=str)[:80]}")

# ── 2. From VynCo company profile (optional) ────────────────────────────────

# Uncomment to try the VynCo path. Requires the portal to have VynCo wired
# up and your tier to include the integration; falls back with 404/503 otherwise.
#
# try:
#     print("\n=== configs.from_company(name=...) ===")
#     co = client.configs.from_company(name="Nestlé", periods=4, fraud_rate=0.02)
#     print(f"company: {co.company.get('name')}  "
#           f"({co.company.get('uid')})")
#     print(f"notes:   {co.notes}")
# except vynfi.NotFoundError:
#     print("\n(VynCo company not found — integration may be disabled in your portal.)")

# ── 3. (Optional) submit the generated config for a small run ────────────────

if os.environ.get("VYNFI_RUN_NL", "") == "1":
    # Bound the dry-run to something small regardless of what the LLM chose.
    cfg = dict(nl.config or {})
    cfg["rows"] = min(int(cfg.get("rows", 200)), 200)
    cfg["periods"] = min(int(cfg.get("periods", 1)), 1)
    cfg["companies"] = min(int(cfg.get("companies", 1)), 1)
    print(f"\nSubmitting NL-generated config (rows={cfg['rows']})…")
    job = client.jobs.generate_config(config=cfg)
    done = client.jobs.wait(job.id, timeout=300)
    print(f"  job {done.id} → {done.status}")