Agentic: Evidence-First Hiring with MCP, AWS Lambda, and DynamoDB Streams
We built Agentic, an evidence-first hiring platform where candidates apply through AI assistants using MCP and recruiters review repository-backed technical evidence. This article explains how AWS SAM, API Gateway, Lambda, DynamoDB Streams, and S3 power the event-driven evaluation workflow.
# Agentic: Evidence-First Hiring with MCP, AWS Lambda, and DynamoDB Streams
Hiring teams often rely on resumes and self-reported technical skills. However, a resume does not always show how a candidate structures code, writes tests, maintains a repository, or approaches real engineering problems.
For the WeMakeDevs AWS First Commit hackathon, we built **Agentic**, an evidence-first hiring platform that helps recruiters review technical evidence from candidate repositories before interviews.
Candidates can apply through the web application or through AI assistants connected using the **Model Context Protocol (MCP)**. After submission, the application enters an AWS-powered evaluation workflow that produces a structured candidate intelligence report for human review.
## The problem
Traditional hiring workflows create challenges for both candidates and recruiters.
Candidates repeatedly fill out similar forms and often struggle to demonstrate the quality of their real work. Recruiters, meanwhile, must review large numbers of resumes and repositories manually. This makes technical screening time-consuming and difficult to standardize.
We wanted to build a workflow where:
- Candidates could apply through their AI assistants.- Applications could include repository-backed evidence.- Repository evaluation could happen asynchronously.- Recruiters could review structured technical findings.- AI recommendations would support human decisions rather than replace them.
## How Agentic works
The workflow begins when a candidate or AI assistant discovers an open role through the MCP server. The assistant can retrieve the job description, required skills, application requirements, and evaluation criteria.
The candidate then submits an application containing their identity, technical skills, project information, and public repository URL. Before submission, the candidate reviews the application and explicitly confirms it.
The application is stored in Amazon DynamoDB. A DynamoDB Stream then notifies the evaluation workflow that a new application has been created.
The evaluator analyzes the repository for technical signals such as:
- Git history and contribution patterns- Software architecture and modularity- Test quality and assertion depth- Security and code-quality concerns- Repository structure- Evaluation events and generated artifacts
The results are made available to recruiters through the Recruiter Control Room.
## Architecture
The main workflow is:
```textCandidate or AI Assistant | vMCP/API Request | vAPI Gateway and Lambda | vDynamoDB Application Record | vDynamoDB Stream | vContainerized Evaluation Workflow | vS3 Reports and Execution Artifacts | vRecruiter Control RoomThis design separates application intake from repository evaluation. The candidate does not need to wait for every evaluation stage to finish before the application is accepted.
AWS services used
AWS SAM
We used AWS Serverless Application Model to define the core backend and evaluation infrastructure as code.
The SAM configuration describes the Lambda functions, DynamoDB tables, DynamoDB Stream event source, S3 bucket, IAM permissions, and API integration.
This allowed us to validate, build, and deploy the backend using:
sam validatesam buildsam deployUsing infrastructure as code made it easier to reproduce the architecture and iterate on the deployment configuration.
Amazon API Gateway
Amazon API Gateway exposes the HTTP APIs used by the recruiter frontend and MCP integration.
This allows external AI assistants and frontend clients to interact with the hiring workflow through standardized endpoints.
AWS Lambda
AWS Lambda powers the recruiter API and the asynchronous evaluation workflow.
The evaluator is packaged as a container image so that it can use the dependencies required for repository inspection and report generation.
Amazon DynamoDB
DynamoDB stores job requisitions and candidate applications.
We use indexes to support job discovery, application lookup, and recruiter-facing queries. DynamoDB’s low-latency access makes it suitable for the application and status-tracking workflow.
DynamoDB Streams
DynamoDB Streams are the central event-driven component of the architecture.
When a new application is inserted, the stream provides an event that can trigger the evaluator. This separates the lightweight task of receiving an application from the heavier task of analyzing a repository.
The flow is:
Application submitted | vApplication stored in DynamoDB | vDynamoDB Stream event created | vEvaluator workflow starts | vReport and artifacts storedThis separation makes the system more responsive and provides a foundation for retrying or extending evaluation stages in the future.
Amazon S3
Amazon S3 stores candidate intelligence reports, structured evaluation artifacts, and execution traces.
The bucket is configured with server-side encryption and public-access blocking. Recruiter-facing APIs retrieve the relevant artifacts for display in the control room.
AWS Secrets Manager
AWS Secrets Manager stores the evaluator’s model API credential outside the source code.
This keeps sensitive configuration separate from the application repository and deployment templates.
The Recruiter Control Room
The recruiter experience is designed around evidence rather than a single opaque score.
A candidate dossier can include:
- Candidate summary
- Role match signals
- Technical skills
- Repository information
- Test findings
- Architecture observations
- Security and code-quality findings
- Strengths and concerns
- Suggested technical interview questions
- Evaluation artifacts and execution events
The platform distinguishes between candidate-provided information, repository-backed evidence, and AI-generated recommendations.
The final hiring decision remains with the recruiter. Agentic is intended to improve the quality and speed of technical review, not to make automatic hiring decisions.
Why MCP matters
MCP allows AI assistants to interact with the hiring workflow through defined tools instead of relying on an unstructured conversation.
The applicant MCP server supports workflows such as:
- Searching for open roles
- Retrieving role details
- Validating a candidate profile
- Submitting an application
- Checking application status
This creates a more flexible candidate experience while keeping the application contract structured and predictable.
What we learned
Our biggest learning was how useful event-driven architecture can be for AI workflows.
DynamoDB Streams allowed us to separate fast application intake from heavier repository analysis. This made the candidate experience more responsive and gave the evaluation process room to evolve independently.
We also learned that containerized evaluation workflows require careful attention to dependencies, permissions, timeouts, deployment packaging, and failure handling.
AWS SAM helped us iterate faster by allowing infrastructure and application configuration to live together. At the same time, integrating API Gateway, Lambda, DynamoDB Streams, containerized workloads, and S3 required debugging across multiple AWS services.
Limitations and next steps
The current prototype focuses on the core application, evaluation, and recruiter-review workflow.
Future improvements could include:
- More granular role-specific evaluation policies
- Additional candidate privacy controls
- Stronger authentication and authorization
- More configurable evaluation plugins
- Improved evaluator retry and observability workflows
- More recruiter collaboration and feedback features
- Additional deployment options for larger evaluation workloads
Conclusion
Agentic gives candidates a more transparent way to demonstrate real engineering work and gives recruiters structured technical evidence before an interview.
The project combines MCP with an event-driven AWS architecture so that candidates can interact through AI assistants while recruiters receive reviewable repository evidence.
The goal is not to replace human judgment. The goal is to give humans better information before making important hiring decisions.
Verify the work behind the resume.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article