Analisis 10.000 dokumen dalam semalam.

# enqueue.py — fire one batch request per document.
import os
import json
import sqlite3
import httpx
from pathlib import Path
from schemas import SCHEMA

API = "https://api.anoman.io"
KEY = os.environ["ANOMAN_API_KEY"]

# Local SQLite for progress + idempotency. Reuse this script — already-
# queued docs are skipped on re-run.
db = sqlite3.connect("batch_run.db")
db.execute("""
    CREATE TABLE IF NOT EXISTS jobs (
        doc_path TEXT PRIMARY KEY,
        job_id   TEXT,
        status   TEXT DEFAULT 'queued',
        result   TEXT,
        error    TEXT,
        enqueued_at REAL DEFAULT (strftime('%s', 'now'))
    )
""")
db.commit()

SYSTEM = """Extract structured metadata from the document. Return JSON
matching the provided schema. If a field is not present in the doc,
omit it. Don't hallucinate values."""

def enqueue_doc(doc_path: Path):
    # Skip if already queued
    cur = db.execute("SELECT job_id FROM jobs WHERE doc_path = ?", (str(doc_path),))
    if row := cur.fetchone():
        return row[0]

    content = doc_path.read_text()  # or PDF→text extraction
    response = httpx.post(
        f"{API}/v1/chat/completions",
        headers={
            "Authorization": f"Bearer {KEY}",
            "x-anoman-prefer-batch": "true",
        },
        json={
            "model": "deepseek-v3",  # Long context, batch-friendly, cheap
            "messages": [
                {"role": "system", "content": SYSTEM},
                {"role": "user",   "content": f"Schema:\n{json.dumps(SCHEMA)}\n\nDoc:\n{content[:50_000]}"},
            ],
            "response_format": {"type": "json_object"},
            "metadata": {"doc_path": str(doc_path)},
        },
    )
    response.raise_for_status()
    job = response.json()

    db.execute(
        "INSERT INTO jobs (doc_path, job_id) VALUES (?, ?)",
        (str(doc_path), job["id"]),
    )
    db.commit()
    return job["id"]

# Walk a directory of docs
DOCS = Path("./docs_to_analyze")
for doc in DOCS.glob("*.txt"):  # or *.pdf
    enqueue_doc(doc)
    print(f"queued {doc.name}")

# progress.py — run anytime to see how far along the batch is.
import sqlite3

db = sqlite3.connect("batch_run.db")
total, queued, complete, failed = db.execute("""
    SELECT
        COUNT(*),
        SUM(CASE WHEN status = 'queued'   THEN 1 ELSE 0 END),
        SUM(CASE WHEN status = 'complete' THEN 1 ELSE 0 END),
        SUM(CASE WHEN status = 'failed'   THEN 1 ELSE 0 END)
    FROM jobs
""").fetchone()

print(f"Total:    {total}")
print(f"Queued:   {queued}    ({100*queued/total:.1f}%)")
print(f"Complete: {complete}  ({100*complete/total:.1f}%)")
print(f"Failed:   {failed}    ({100*failed/total:.1f}%)")

# Total savings from batch routing — _anoman.savings_usd on each result
import json
total_saved = 0
for (result_json,) in db.execute("SELECT result FROM jobs WHERE status = 'complete'"):
    body = json.loads(result_json) if result_json else {}
    saved = body.get("_anoman", {}).get("savings_usd", "0")
    total_saved += float(saved)

print(f"Total saved vs realtime: ${total_saved:.2f}")