Inspiration
Hiring is still heavily manual. Recruiters must read hundreds of resumes, coordinate interviews, evaluate candidates, and update Applicant Tracking Systems (ATS). This process is slow, repetitive, and prone to bias. We wanted to explore how agentic AI systems could automate complex recruiting workflows while keeping decisions structured and explainable.
Our goal was to build an autonomous workflow where multiple AI agents collaborate to evaluate candidates, make hiring decisions, and simulate recruiting operations.
What it does
NovaOps Agent is a multi-agent AI recruiting workflow automation system.
The platform analyzes job descriptions and resumes, performs candidate screening, generates interview questions, evaluates responses, and produces structured hiring recommendations.
The system is composed of specialized agents coordinated by a workflow manager:
Planner Agent Generates the workflow plan for evaluating candidates.
Screening Agent Uses Amazon Nova 2 Lite via Amazon Bedrock to analyze resumes against job requirements and produce structured evaluation results.
Interview Agent Simulates a technical interview by generating and collecting candidate responses.
Evaluation Agent Applies weighted scoring across multiple criteria such as Python expertise, system design ability, and communication skills.
Execution Agent Simulates ATS actions such as updating candidate status and triggering hiring workflows.
Verification Agent Validates that workflow actions completed successfully and generates the final report.
The system outputs a structured hiring decision including scores, reasoning, and recommendation.
How we built it
NovaOps Agent is implemented as a modular multi-agent architecture in Python.
The system uses a centralized Workflow Manager that orchestrates specialized agents in sequence:
Planner → Screening → Interview → Evaluation → Execution → Verification
Key components include:
Amazon Bedrock for model access
Amazon Nova 2 Lite for candidate reasoning and resume evaluation
Structured JSON pipelines for communication between agents
Modular agent classes responsible for distinct tasks
Logging utilities to trace workflow execution and decision points
The architecture is designed to be extensible, allowing additional agents or automation tools to be integrated easily.
Challenges we ran into
One of the biggest challenges was integrating the Nova model into the pipeline while maintaining structured outputs.
LLM responses often include extra formatting or explanation text, so we needed to design prompts and validation logic that ensured the model returned clean JSON that the agents could reliably parse.
Another challenge was designing a modular architecture where agents could operate independently while still passing structured data between stages of the workflow.
Finally, configuring AWS authentication and Bedrock access required careful setup to ensure the agents could securely invoke the Nova model.
Accomplishments that we're proud of
We successfully built a fully functioning multi-agent AI system that executes a recruiting workflow from start to finish.
Key accomplishments include:
Integrating Amazon Nova reasoning into an automated hiring pipeline
Designing a modular multi-agent architecture
Building a working orchestration system coordinating multiple agents
Producing structured hiring decisions with explainable scoring
Simulating ATS workflow automation
The final system demonstrates how agentic AI can automate complex business workflows.
What we learned
Through building NovaOps Agent we learned how powerful agent-based architectures can be when combined with modern foundation models.
We also learned the importance of:
structured prompts for reliable model outputs
validation layers to ensure system robustness
modular agent design to simplify orchestration and debugging
Using Nova through Amazon Bedrock showed how enterprise-grade AI models can be integrated into real-world workflow automation systems.
What's next for Novaops Agent
Future versions of NovaOps Agent could expand into a full recruiting automation platform.
Potential next steps include:
Integrating Nova-powered Planner Agents to dynamically design interview workflows
Adding real ATS integrations such as Greenhouse or Lever
Supporting voice-based interviews using speech AI
Adding dashboards for recruiters to monitor workflow execution
Using embeddings to improve candidate-job matching at scale
Ultimately, we envision NovaOps Agent evolving into an AI-powered system that helps organizations recruit faster, more fairly, and more efficiently.
Log in or sign up for Devpost to join the conversation.