Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/vrashmanyu605-eng/Langchain_Interview_Multi_Agents_Flow/llms.txt

Use this file to discover all available pages before exploring further.

The evaluation agent is intentionally called last. It waits until all six of its required inputs are present in state before the supervisor routes to it. At that point it has a complete picture of the candidate: the parsed profile, the fit score, the research insights, and all three interview question banks. It synthesizes everything into a single structured output — the evaluation field — that the email agent then uses to draft the interviewer briefing emails.

​
Source code

from llm import llm

def evaluation_agent(state):

    candidate_profile = state["candidate_profile"]
    matching_analysis = state["matching_analysis"]
    research_analysis = state["research_analysis"]
    hr_questions = state["hr_questions"]
    technical_questions = state["technical_questions"]
    ceo_questions = state["ceo_questions"]

    response = llm.invoke(
        f"""
        You are a Hiring Evaluation Agent.

        Generate a final hiring evaluation.

        Candidate:
        {candidate_profile}

        Matching Analysis:
        {matching_analysis}

        Research Analysis:
        {research_analysis}

        HR Questions:
        {hr_questions}

        Technical Questions:
        {technical_questions}

        CEO Questions:
        {ceo_questions}

        Generate:
        - overall recommendation
        - confidence score
        - strengths
        - concerns
        - hiring risks
        - final verdict

        Return structured JSON only.
        """
    )

    return {
        "evaluation": response.content
    }

​
Inputs

All six fields must be present in state before the supervisor routes to this agent. Note that the supervisor’s agents_info registry lists only five requires fields (matching_analysis, research_analysis, hr_questions, technical_questions, ceo_questions) — candidate_profile is omitted from the registry but is still read directly by the agent function. In practice, candidate_profile is always available before evaluation is called since it is one of the first fields produced.
​
candidate_profile
string
required
The structured JSON profile produced by the resume parser agent.
​
matching_analysis
string
required
The fit score, strengths, weaknesses, and recommendation produced by the matching agent.
​
research_analysis
string
required
The strategic interview insights produced by the candidate research agent.
​
hr_questions
string
required
The 15-question HR bank with evaluation criteria and red flags.
​
technical_questions
string
required
The 20-question technical bank with expected answers, follow-ups, and difficulty ratings.
​
ceo_questions
string
required
The 10-question CEO bank with leadership criteria, behavioral indicators, and red flags.

​
Output

​
evaluation
string
A JSON string containing the final hiring recommendation. Read by the email agent to generate the interviewer briefing emails.
The JSON contains the following fields:
FieldDescription
overall_recommendationA label such as “Strong Hire”, “Hire”, “No Hire”, or “Strong No Hire”
confidence_scoreA numeric score (0–100) reflecting how confident the LLM is in the recommendation
strengthsA list of the candidate’s standout qualities relative to the role
concernsAreas where the candidate falls short of the role’s expectations
hiring_risksSpecific risks that could affect the candidate’s success in the role
final_verdictA prose summary of the recommendation with actionable next steps

​
Example output

{
  "overall_recommendation": "Strong Hire",
  "confidence_score": 82,
  "strengths": ["Deep Python expertise", "Solid system design foundation"],
  "concerns": ["Limited cloud-native experience"],
  "hiring_risks": ["May require ramp-up on Kubernetes"],
  "final_verdict": "Recommend proceeding with offer subject to successful technical round"
}

​
State diagram

Requires: candidate_profile, matching_analysis, research_analysis,
          hr_questions, technical_questions, ceo_questions
    ↓
evaluation_agent
    ↓
Provides: evaluation
The evaluation agent requires all six inputs to be present in state before it is called. The supervisor’s input validation ensures this.

Build docs developers (and LLMs) love

Get started for freeTalk to us