feder-cr/Jobs_Applier_AI_Agent_AIHawk · warning · ValueError

No numbers found in the string

Error message

No numbers found in the string

What it means

extract_number_from_string pulls all digit runs out of the LLM's answer with re.findall(r'\d+'). If the model answered a numeric question with no digits at all (e.g. 'N/A' or prose), the function raises ValueError because it cannot return a number.

Source

Thrown at src/libs/llm_manager.py:647

        try:
            output = self.extract_number_from_string(output_str)
            logger.debug(f"Extracted number: {output}")
        except ValueError:
            logger.warning(
                f"Failed to extract number, using default experience: {default_experience}"
            )
            output = default_experience
        return output

    def extract_number_from_string(self, output_str):
        logger.debug(f"Extracting number from string: {output_str}")
        numbers = re.findall(r"\d+", output_str)
        if numbers:
            logger.debug(f"Numbers found: {numbers}")
            return str(numbers[0])
        else:
            logger.error("No numbers found in the string")
            raise ValueError("No numbers found in the string")

    def answer_question_from_options(self, question: str, options: list[str]) -> str:
        logger.debug(f"Answering question from options: {question}")
        func_template = self._preprocess_template_string(prompts.options_template)
        prompt = ChatPromptTemplate.from_template(func_template)
        chain = prompt | self.llm_cheap | StrOutputParser()
        raw_output_str = chain.invoke(
            {
                RESUME: self.resume,
                JOB_APPLICATION_PROFILE: self.job_application_profile,
                QUESTION: question,
                OPTIONS: options,
            }
        )
        output_str = self._clean_llm_output(raw_output_str)
        logger.debug(f"Raw output for options question: {output_str}")
        best_option = self.find_best_match(output_str, options)
        logger.debug(f"Best option determined: {best_option}")

View on GitHub (pinned to 79155b52fa)

Solutions

1. Retry the numeric question: instruct the model to answer with digits only; often the second attempt works.
2. Use a stronger model or lower temperature for the numeric chain.
3. Wrap the call and default to '0' or skip the question when the model gives no numeric answer.

Example fix

# before
val = llm_manager.extract_number_from_string(ans)
# after
try:
    val = llm_manager.extract_number_from_string(ans)
except ValueError:
    val = '0'  # or re-invoke the numeric chain with a digits-only instruction

import re
if not re.search(r'\d', llm_answer):
    llm_answer = '0'  # or re-prompt with 'answer with digits only'

def contains_digit(s: str) -> bool:
    return bool(re.search(r'\d', s))

try:
    val = llm_manager.extract_number_from_string(ans)
except ValueError:
    val = '0'  # or retry the numeric chain once

When it happens

Trigger: answer_question_numeric invokes the LLM for a numeric field (years of experience, salary) and the reply contains no digits, so numbers is empty and the error is raised.

Common situations: LLM refusing or hedging on numeric questions ('I prefer not to say'), non-English digit formats being unlikely but possible (e.g. spelled-out numbers), or a weak/cheap model ignoring the instruction to answer with a number.

Related errors

• Could not extract section name from the response.
• Unsupported model type: {llm_model_type}
• Section '{section_name}' not found in either resume or job_a
• Chain not defined for section '{section_name}'
• Failed to get a response from the model after multiple attem

AI-assisted analysis of feder-cr/Jobs_Applier_AI_Agent_AIHawk@79155b52fa (2026-08-28). Data as JSON: /api/errors/37c1cc74a9e385ee. Report an issue: GitHub.