The Problem
Grading 100+ IT practical assessments by hand takes hours — checking Delphi code for correctness, verifying SQL queries produce the right output, and evaluating theory answers against CAPS rubrics. With three grades (10, 11, 12) and multiple classes, the marking load was unsustainable.
The Solution: MaaS
MaaS (Marking-as-a-Service) is a multi-agent pipeline that:
- Parses student submissions (Delphi
.pasfiles, SQL scripts, and theory answer sheets) - Evaluates each component against CAPS rubrics using specialized marking agents
- Aggregates results into score reports with detailed feedback
Architecture
The system uses Hermes Agent to orchestrate parallel sub-agents, each specialized for a specific marking task:
- Practical Marker — evaluates Delphi code for correctness, efficiency, and style
- SQL Marker — runs queries against test databases and checks output
- Theory Marker — assesses written answers against keyword-based rubrics
- Batch Controller — distributes work across multiple assessments and merges results
Key Results
- Marking time reduced from ~8 hours to ~30 minutes per batch
- Consistent, rubric-aligned grading across all submissions
- Detailed feedback automatically generated for each student
- Teachers can focus on teaching instead of grading
Tech Stack
- Orchestration: Hermes Agent with sub-agent delegation
- Languages: Python, Delphi, SQL
- Infrastructure: Ubuntu VPS, Docker, systemd services
- Deployment: Git-based CI/CD with automated testing
This project continues to evolve — next up: integrating with the school’s LMS for direct grade posting.