paper_firehose.processors.pdf_summarizer
Full-text PDF summarization with a single OpenAI call per paper.
The paper’s text is extracted locally with pypdf and sent, together with the
topic’s prompt, to the OpenAI Responses API. Structured outputs (a strict JSON
schema) guarantee a {"summary": ..., "methods": ...} object, which is the
shape the HTML and email renderers expect in paper_qa_summary.
Functions
build_instructions(prompt, ranking_query)
|
Combine the base instructions with the topic prompt ({ranking_query} substituted). |
extract_pdf_text(pdf_path, *[, max_chars])
|
Extract text from a PDF, truncated to max_chars. |
Classes
PdfSummarizer([model, reasoning_effort, client])
|
Summarize paper full text with one OpenAI Responses API call. |
-
class paper_firehose.processors.pdf_summarizer.PdfSummarizer(model=None, *, reasoning_effort=None, client=None)[source]
Bases: object
Summarize paper full text with one OpenAI Responses API call.
- Parameters:
-
-
summarize_pdf(pdf_path, instructions)[source]
Extract the PDF’s text and summarize it.
- Return type:
Optional[Dict[str, str]]
- Parameters:
pdf_path (str)
instructions (str)
-
summarize_text(text, instructions)[source]
Return {"summary", "methods"} for the given paper text, or None on failure.
- Return type:
Optional[Dict[str, str]]
- Parameters:
-
-
paper_firehose.processors.pdf_summarizer.build_instructions(prompt, ranking_query)[source]
Combine the base instructions with the topic prompt ({ranking_query} substituted).
- Return type:
str
- Parameters:
-
Extract text from a PDF, truncated to max_chars. Returns ‘’ on failure.
- Return type:
str
- Parameters:
pdf_path (str)
max_chars (int)