Analyze Single Paper¶
Start with one source so you can tune prompt quality before scaling.
The goal is a quality baseline: one file, one prompt, one answer you can inspect manually. Get this right first — everything else builds on it.
Run It¶
python -m cookbook getting-started/analyze-single-paper \
--input cookbook/data/demo/text-medium/input.txt --mock
With a real API:
python -m cookbook getting-started/analyze-single-paper \
--input path/to/file.pdf --no-mock --provider gemini --model gemini-2.5-flash-lite
What You'll See¶
Status: ok
Answer (excerpt): The paper presents three main findings: (1) context caching
reduces repeated token cost by 85-92%, (2) fan-out patterns benefit most from
caching, and (3) broadcast execution scales linearly with source count...
Tokens: 1,340 (prompt: 1,250 / completion: 90)
Status: ok means the request path is healthy. The excerpt should be specific
to your source — not generic boilerplate. Token usage gives your first
cost-per-document estimate.
Tuning¶
- Use
--promptto tighten format requirements (bullets, table, JSON). - Keep this recipe as a golden baseline before changing models or scaling.
- If the output is vague, make task constraints explicit in the prompt.
Next Steps¶
Once your single-source output looks good, scale to multiple files with Broadcast Process Files.