Skip to main content

​
The FairMatch evaluation process

FairMatch uses a sophisticated multi-agent AI system to evaluate candidates objectively. Each agent specializes in a specific evaluation dimension, and the orchestrator coordinates their work to produce a comprehensive assessment.
1

Candidate submission

Candidates upload their resume (PDF) and complete an optional AI interview questionnaire. Companies can also bulk import candidates from Excel/CSV files.
2

Data extraction

The Resume Analyst agent extracts structured data from the uploaded documents:
  • Name and contact information
  • Skills and technical expertise
  • Years of experience
  • Project descriptions
def get_resume_intelligence(resume_text: str) -> dict:
    # Query the Resume Analysis agent
    agents = orchestrator.search_agents_by_keyword("FairMatch Resume Analyst")
    msg = AgentMessage(content=resume_text, sender_id=orchestrator.agent_id)
    response = orchestrator.x402_processor.post(sync_url, json=msg.to_dict())
    return json.loads(response.json().get('response', '{}'))
3

Parallel agent evaluation

Five specialized agents evaluate the candidate simultaneously:1. Resume Analyst - Extracts and structures candidate data2. GitHub Verifier - Validates technical contributions and project quality3. Interview Grader - Assesses interview responses for depth and accuracy4. Decision Intelligence - Analyzes skill match, experience alignment, and consistency5. Integrity Analyst - Detects potential fraud or resume fabrication
4

Adaptive fraud detection

If GitHub and interview scores differ by more than 30 points, the system automatically spawns the Integrity Analyst:
if abs(github_score - interview_score) > 30 and candidate.github_link:
    print(f"FRAUD DETECTED: Spawning Integrity Agent...")
    fraud_query = f"GitHub Score: {github_score}\nInterview Score: {interview_score}"
    integrity_json = get_integrity_intelligence(fraud_query)
    integrity_penalty = integrity_json.get("penalty_score", 0)
5

Blind mode evaluation

During evaluation, all PII is masked. Evaluators see candidates as “Candidate ID-1234” instead of seeing names, emails, or demographic information.
6

Weighted scoring

The final score is calculated using company-defined weights:
final_score = (
    skill_score * (job.weight_skill / total_weight) +
    github_score * (job.weight_github / total_weight) +
    interview_score * (job.weight_interview / total_weight) +
    experience_score * (job.weight_experience / total_weight) +
    integrity_score * (job.weight_integrity / total_weight)
)

​
Agent orchestration architecture

FairMatch uses the Zynd AI agent framework to coordinate specialized agents. The orchestrator handles:
  • Agent discovery via the Zynd registry
  • Message routing between agents
  • Payment processing for agent calls (x402 protocol)
  • Response aggregation and error handling
agent_config = AgentConfig(
    name="FairMatch Orchestrator",
    description="Orchestrates FairMatch AI evaluations",
    capabilities={"ai": ["orchestration"], "protocols": ["http"]},
    webhook_host="0.0.0.0",
    webhook_port=5000,
    registry_url="https://registry.zynd.ai"
)

orchestrator = ZyndAIAgent(agent_config=agent_config)
Each agent call is automatically paid for using the x402 payment protocol, enabling a decentralized marketplace of AI evaluation services.

​
Data flow diagram

​
Evaluation result structure

Each candidate receives a comprehensive evaluation with multiple dimensions: 
class EvaluationResult:
    candidate_id: str
    job_id: str
    name: str
    skill_score: int        # 0-100: Skill match analysis
    github_score: int       # 0-100: Code quality and contributions
    interview_score: int    # 0-100: Interview response quality
    experience_score: int   # 0-100: Years of experience alignment
    integrity_score: int    # 0-100: Consistency and authenticity
    final_score: int        # Weighted average of all scores
    strengths: List[str]    # AI-identified candidate strengths
    weaknesses: List[str]   # AI-identified risk factors
    risk_level: str         # low, medium, high
    recommendation: str     # consider, interview, reject
All scores are calculated objectively by AI agents with no access to candidate demographics.

​
Company workflow

  1. Create a job listing with required skills and experience
  2. Upload candidates via CSV/Excel or accept individual applications
  3. Run evaluation to trigger the multi-agent pipeline
  4. Toggle blind mode to review candidates without bias
  5. Review rankings in a secure dashboard with radar charts
  6. Make hiring decisions based on objective scores

​
Candidate workflow

  1. Register an account on the FairMatch platform
  2. Browse open jobs from companies using FairMatch
  3. Upload your resume (PDF) to auto-fill your application
  4. Complete the AI interview questionnaire (optional)
  5. Track your applications in your candidate dashboard
Candidates should ensure their GitHub profiles are public and up-to-date, as the GitHub Verifier agent will analyze commit history and project quality.

Build docs developers (and LLMs) love

Get started for freeTalk to us