A standard-compliant README for the AI Grading Automation System, a multi-agent, human-in-the-loop hyperautomation solution designed for educational institutions.
This repository contains the version-controlled backup of all n8n automation workflows that power the AI-enabled grading system used by university instructors. These workflows orchestrate a full evaluation pipeline integrating multi-agent reasoning, Google Workspace, Gemini/GPT, Supabase, and advanced hyperautomation lifecycle (HAL) patterns.
The system dramatically reduces grading time, standardizes evaluation criteria, and augments the instructor with intelligent, auditable decision support.
University instructors spend 6 to 8 hours manually grading a single group of assignments. This produces:
- Delayed feedback cycles
- Inconsistent scoring due to fatigue
- Lack of standardized rubrics
- Limited visibility and traceability
- High operational burden
The AI Grading Automation System solves this by implementing a multi-agent orchestration layer that automatically:
- Generates rubrics using pedagogical agents
- Classifies and preprocesses student deliverables
- Executes LLM-based evaluations
- Produces personalized feedback
- Logs grades automatically
- Routes results for human approval before delivery
This repository preserves all the n8n workflows that enable the system.
To run these workflows, you need:
-
A deployed n8n instance (Docker, cloud container, or SaaS)
-
API credentials for:
- Google Drive / Google Sheets
- Gmail or Outlook
- Gemini or GPT
- Supabase
Clone this backup repository:
git clone https://github.com/LePeanutButter/ai-grading-automation-workflows-backup.git
cd ai-grading-automation-workflows-backupImport the JSON workflows through the n8n UI.
This repository is strictly a workflow backup, ready to be imported into any n8n instance. Once imported, the automations will:
- Detect new assignments in Google Drive
- Retrieve or generate the appropriate rubric
- Preprocess and analyze the content
- Evaluate with LLMs through agentic workflows
- Request human approval (HITL)
- Email personalized feedback
- Log results in Google Sheets
The system implements a distributed multi-agent architecture on top of n8n, where each workflow component embodies a functional agent with a defined responsibility.
| Agent | Responsibility |
|---|---|
| Rubric Agent | Generates and adapts rubrics using LLMs and pedagogical frameworks |
| Evaluation Agent | Analyzes submissions and produces scoring + personalized feedback |
| Document Agent | Handles preprocessing: OCR, parsing, format validation |
| Memory Agent | Stores rubric history, instructor preferences, course context |
| Compliance Agent | Enforces security, anonymization, and audit requirements |
| Feedback Delivery Agent | Sends personalized emails to students |
| Instructor Approval Agent | Manages HITL checkpoints |
This multi-agent structure supports scalability, modularity, and clear separation of responsibilities, aligning with modern agentic design best practices.
The system is intentionally not fully autonomous. It embeds a strict HITL safeguard:
- Every AI-generated evaluation must be reviewed, corrected, or approved by the instructor.
- Feedback is only delivered after explicit human confirmation.
- All HITL interventions are logged for audit, version control, and fairness guarantees.
This creates a hybrid workflow that enhances instructor productivity while maintaining academic integrity.
The system includes a memory layer managed through Supabase and internal n8n storage:
- Saves rubric versions
- Tracks instructor preferences
- Stores historical submissions
- Maintains course-level context
- Logs evaluation patterns for consistency
This memory enables more consistent evaluations and reduces repetitive tasks across assignments, semesters, and courses.
The system implements the Hyperautomation Lifecycle (HAL) end-to-end:
-
Discover: Identify bottlenecks: grading, rubric creation, feedback delays.
-
Analyze: Process breakdown, stakeholder interviews, latency assessment.
-
Design: Multi-agent architecture + rubric generation model + HITL checkpoints.
-
Automate: n8n workflows, LLM pipelines, and automatic data routing.
-
Orchestrate: Cross-service coordination via Google Workspace, Supabase, email pipelines.
-
Optimize: Metrics dashboards (Grafana), accuracy tuning, cost reduction.
-
Govern: Audit logs, access control, security policies, compliance oversight.
This positions the solution as a full hyperautomation platform, not just an LLM tool.
Security was a first-class requirement throughout the system design.
- All secrets stored via n8n Credential Manager, encrypted at rest.
- No credentials embedded in workflows.
- Deliverables are anonymized before LLM processing.
- Sensitive metadata is redacted.
-
All communications use TLS:
- n8n -> Google APIs
- n8n -> Email services
- n8n -> Supabase
- Instructors authenticate through secure channels.
- Only authorized users can approve or release feedback.
-
Every evaluation, modification, decision, and data access is logged:
- Timestamp
- Instructor identity
- AI evaluation
- Final feedback
This aligns with institutional requirements for educational data protection.
The automation system supports two commercialization strategies:
Individual instructors can purchase grading automation as a service.
- Fast onboarding
- Low cost per workflow
- Individual academic use
- Immediate ROI (pays for itself after grading one group)
Universities can integrate the system at scale.
- Multi-course and multi-semester deployments
- Centralized dashboards for academic analytics
- Integration with LMS systems (Canvas, Moodle, Blackboard)
- Governance + audit layers for administrative oversight
- Volume-based pricing
The combination of agent automation, HAL governance, and data security strengthens the B2B value proposition and makes the solution scalable at institutional levels.
- Jcro15 - Juan Camilo Rojas Ortiz
- lrvalencia - Luiggi Valencia Vélez
- LePeanutButter - Santiago Botero
MIT © Juan Camilo Rojas Ortiz, Luiggi Valencia Vélez, Santiago Botero García
This README follows the Standard Readme specification.