"""OpenAI API client wrapper for cross-model evaluation.""" from __future__ import annotations import base64 import os import httpx from dotenv import load_dotenv from openai import AsyncOpenAI load_dotenv() _client: AsyncOpenAI | None = None DEFAULT_MODEL = "gpt-5.2" def get_client() -> AsyncOpenAI: global _client if _client is None: api_key = os.environ.get("OPENAI_API_KEY") if not api_key: raise RuntimeError("OPENAI_API_KEY environment variable not set") _client = AsyncOpenAI(api_key=api_key) return _client async def assess_with_text( text: str, system_prompt: str, user_prompt: str, model: str = DEFAULT_MODEL, temperature: float = 0, max_tokens: int = 65536, ) -> str: """Send an assessment request with manuscript text (no PDF) to OpenAI.""" client = get_client() response = await client.chat.completions.create( model=model, temperature=temperature, max_completion_tokens=max_tokens, messages=[ {"role": "system", "content": system_prompt}, { "role": "user", "content": f"\n{text}\n\n\n{user_prompt}", }, ], ) return response.choices[0].message.content or "" async def download_pdf(pdf_url: str) -> bytes: """Download a PDF from a URL and return raw bytes.""" async with httpx.AsyncClient(follow_redirects=True, timeout=60) as client: resp = await client.get(pdf_url) resp.raise_for_status() return resp.content async def assess_with_pdf( pdf_bytes: bytes, system_prompt: str, user_prompt: str, model: str = DEFAULT_MODEL, temperature: float = 0, max_tokens: int = 65536, ) -> str: """Send an assessment request with a PDF attachment to OpenAI. Uses base64-encoded PDF as a file content part in the messages API. Returns the raw text response. """ client = get_client() pdf_b64 = base64.standard_b64encode(pdf_bytes).decode("ascii") response = await client.chat.completions.create( model=model, temperature=temperature, max_completion_tokens=max_tokens, messages=[ {"role": "system", "content": system_prompt}, { "role": "user", "content": [ { "type": "file", "file": { "filename": "paper.pdf", "file_data": f"data:application/pdf;base64,{pdf_b64}", }, }, { "type": "text", "text": user_prompt, }, ], }, ], ) return response.choices[0].message.content or "" async def assess_with_pdf_url( pdf_url: str, system_prompt: str, user_prompt: str, model: str = DEFAULT_MODEL, temperature: float = 0, max_tokens: int = 65536, ) -> str: """Send an assessment request with a PDF URL to OpenAI Responses API. Uses file_url instead of base64, bypassing the upload size limit. Returns the raw text response. """ client = get_client() input_messages = [] if system_prompt: input_messages.append({"role": "system", "content": system_prompt}) input_messages.append({ "role": "user", "content": [ { "type": "input_file", "file_url": pdf_url, }, { "type": "input_text", "text": user_prompt, }, ], }) response = await client.responses.create( model=model, temperature=temperature, max_output_tokens=max_tokens, input=input_messages, ) return response.output_text or "" async def assess_with_pdf_images( pdf_bytes: bytes, system_prompt: str, user_prompt: str, model: str = DEFAULT_MODEL, temperature: float = 0, max_tokens: int = 65536, ) -> str: """Send a PDF as page images to OpenAI, bypassing text token limits. Renders each page as a PNG with PyMuPDF and sends as image_url parts. Each image costs ~1k tokens regardless of text content. """ import fitz # PyMuPDF doc = fitz.open(stream=pdf_bytes, filetype="pdf") # Cap at 50 pages to stay within API limits; skip trailing appendix pages max_pages = min(len(doc), 50) image_parts = [] for i in range(max_pages): page = doc[i] pix = page.get_pixmap(dpi=150) png_bytes = pix.tobytes("png") img_b64 = base64.standard_b64encode(png_bytes).decode("ascii") image_parts.append({ "type": "image_url", "image_url": { "url": f"data:image/png;base64,{img_b64}", "detail": "high", }, }) doc.close() client = get_client() content = image_parts + [{"type": "text", "text": user_prompt}] response = await client.chat.completions.create( model=model, temperature=temperature, max_completion_tokens=max_tokens, messages=[ {"role": "system", "content": system_prompt}, {"role": "user", "content": content}, ], ) return response.choices[0].message.content or ""