wehub-resource-sync
73 lines
2.3 KiB
Python
73 lines
2.3 KiB
Python
from datetime import datetime
|
|
from typing import Any, Optional
|
|
|
|
from llmai.shared import JSONSchemaResponse, SystemMessage, UserMessage
|
|
|
|
from utils.llm_utils import generate_structured_with_schema_retries
|
|
|
|
SEARCH_QUERY_GENERATION_PROMPT = """
|
|
Generate a concise web search query that finds useful factual context for a presentation.
|
|
|
|
# Steps
|
|
1. Analyze CONTENT and INSTRUCTIONS to identify the presentation topic and information needs.
|
|
2. Consider today's date when the request needs current or time-sensitive information.
|
|
3. Return one search-engine-style query that would add useful factual context.
|
|
|
|
# Rules
|
|
- Keep the query focused and at most 12 words.
|
|
- Preserve important names, places, dates, products, and technical terms.
|
|
- Include terms such as "latest", the relevant year, statistics, trends, or examples when
|
|
they would improve the presentation.
|
|
- Do not answer the user's request.
|
|
- Do not include explanations, quotation marks, search operators, or multiple queries.
|
|
""".strip()
|
|
|
|
SEARCH_QUERY_RESPONSE_SCHEMA = {
|
|
"type": "object",
|
|
"properties": {
|
|
"query": {
|
|
"type": "string",
|
|
"minLength": 1,
|
|
"maxLength": 200,
|
|
},
|
|
},
|
|
"required": ["query"],
|
|
"additionalProperties": False,
|
|
}
|
|
|
|
|
|
async def generate_web_search_query(
|
|
client: Any,
|
|
model: str,
|
|
content: str,
|
|
instructions: Optional[str] = None,
|
|
) -> Optional[str]:
|
|
response_format = JSONSchemaResponse(
|
|
name="web_search_query",
|
|
json_schema=SEARCH_QUERY_RESPONSE_SCHEMA,
|
|
strict=False,
|
|
)
|
|
response = await generate_structured_with_schema_retries(
|
|
client,
|
|
model,
|
|
messages=[
|
|
SystemMessage(content=SEARCH_QUERY_GENERATION_PROMPT),
|
|
UserMessage(
|
|
content=(
|
|
f"TODAY'S DATE: {datetime.now().strftime('%Y-%m-%d')}\n\n"
|
|
f"CONTENT: {content}\n\n"
|
|
f"INSTRUCTIONS: {instructions or ''}"
|
|
)
|
|
),
|
|
],
|
|
response_format=response_format,
|
|
json_schema=SEARCH_QUERY_RESPONSE_SCHEMA,
|
|
strict=False,
|
|
validate_schema=True,
|
|
)
|
|
query = response.get("query")
|
|
if not isinstance(query, str):
|
|
return None
|
|
normalized_query = " ".join(query.split())[:200]
|
|
return normalized_query or None
|