Hire me Club — Agentic AI Candidate Screening

Mist Avinya designed and deployed a fully serverless Agentic AI recruitment platform on AWS, enabling Hire me Club to autonomously screen, score, and rank candidates against job descriptions — cutting time-to-shortlist from days to hours using Amazon Bedrock Agents and Claude 3.5 Sonnet.

94%Reduction in time-to-shortlist

75%Reduction in recruiter screening time

3.2xPeak hiring volume supported

About the Customer

Business Context & Challenges

Business Context

With a lean team processing 800–1,200 CVs per open role, Hire me Club needed to scale screening capacity without proportionally growing headcount. Manual screening consumed 60–70% of recruiters' working hours.

Key Challenges

Recruiter burnout from repetitive screening · 5–7 day time-to-shortlist losing top candidates · inconsistent evaluation quality · inability to scale during peak seasons · growing SLA pressure.

Solution

How the Agent Works

Agent Workflow

Retrieves job description, skills, and scoring rubric from DynamoDB · fetches CV keys from S3 · extracts text via Textract · evaluates with Claude 3.5 Sonnet using chain-of-thought scoring · writes ranked results back · notifies recruiters via SNS.

Agent Tools (Lambda-backed)

get_job_description · list_candidates · extract_cv_text (async Textract) · submit_candidate_score (persists + notifies). ReAct reasoning loop with per-candidate tool calls, reflection, and human-in-the-loop review for low-confidence candidates.

Foundation Model

Why Claude 3.5 Sonnet

Selection Rationale

Best-in-class tool-use reliability and multi-step reasoning · 200K-token context window handles long CVs and detailed JDs in one pass · native Bedrock integration keeps data inside the AWS account.

Alternatives Considered

Amazon Titan Text — lower reasoning quality for nuanced evaluation. GPT-4o — not AWS-native, data-residency risk. Mistral Large — inconsistent tool-call behaviour in agentic loops.

Security

Security & Responsible AI

Security Architecture

Private VPC with VPC endpoints · least-privilege IAM with temporary STS credentials · KMS encryption at rest, TLS 1.2+ in transit · CloudTrail across all regions · Cognito with MFA and SAML 2.0 federation.

Responsible AI Controls

Bedrock Guardrails strip candidate PII before scoring · low-confidence candidates routed to human review · every score includes a chain-of-thought justification · bias monitoring with SageMaker Clarify · per-posting session isolation.

Business Outcomes

Results at a Glance

Time-to-Shortlist

5–7 business days → 4 hours average (94% reduction)

Recruiter Screening Time

65% of working hours → 16% (75% reduction)

Agent-to-Recruiter Alignment

Manual process → 93% alignment (target exceeded)

Peak Hiring Capacity

1x capacity → 3.2x capacity, no added headcount

Architecture

High Availability & Scalability

Bedrock Agents, Lambda, S3, DynamoDB, SNS, Textract, and API Gateway are inherently multi-AZ. RTO 1 hour, RPO 15 minutes via DynamoDB PITR and S3 versioning. Step Functions fan-out cuts high-volume processing time by 68%.

Have a workflow worth automating?

Tell us what you're working on. We'll share how we'd approach it.

Start a Conversation

← Back to Agentic AI