LIVE ON THE MICROSOFT COMMERCIAL MARKETPLACE  ·  ICO-REGISTERED  ·  UK SOUTH  ·  DPIA-READY

AI Marking Agent — ExComS.ai

AI Marking Agent

Your marking, calibrated to your standards.

ExComS Marking Agent grades student submissions against your mark scheme, learns your personal marking standards over time, and gives you a transparent record of every score and rationale. Built for UK universities, deployed on Azure UK South.

It’s not a generic AI marker. It’s your marker, calibrated to your judgement.

Use cases

Three ways institutions use it

1
Bulk first-pass marking

Upload 200 student submissions in one batch. The Agent grades each one against your mark scheme — per-criterion scores, overall mark, and a written justification — in the time it takes to make a coffee. You correct what needs correcting. The Agent learns from every correction.

2
Calibration across a marking team

When five markers grade the same paper differently, your students notice. Run the Agent across your team’s marking history; it surfaces criterion-level inconsistency, suggests calibration adjustments, and gives the team a shared reference point.

3
Transparent second marker

Use the Agent as a transparent second marker — flagged disagreements, evidence per criterion, audit trail per paper. External examiners can read the rationale without opening the marker’s brain. Disagreements surface for human review before any score is published.

How it works

The calibration loop

Six steps from first upload to a calibrated marking partner.

1
Upload

Mark scheme, manual or PDF.

2
Submit

Bulk upload of papers.

3
Grade

Per criterion, every paper.

4
Correct

You adjust where needed.

5
Learn

Standards lock in.

6
Converge

Gap closes per batch.

Azure OpenAI UK South
Your data stays in the UK. No transfer outside the region.
Temperature = 0
Every paper graded deterministically. Identical input gives identical output.
Audit trail per paper
Every score, rationale, and correction logged. Comply with academic integrity standards.
No public-model retraining
Your data, your students, your IP. We do not feed it back into the foundation model.
How fast, how much

Three concrete savings.

Directional ratios from product capability and pilot data. Verified per-institution figures coming as we close pilot data with our partners.

Time per paper

From 30 minutes of marker time to under 2. Bulk batches of 30 papers grade in the time it takes to make a coffee.

Calibration cost

A new marker takes weeks to calibrate to your team’s standards. The Agent calibrates from your first batch of corrections and converges within 3 batches.

Audit overhead

Every score, rationale, and correction is logged automatically. Audit trails that take days to compile by hand are exported in seconds.

Manual marking vs. AI Marking Agent

Metric Manual marking With AI Marking Agent
Time per paper ~30 minutes Under 2 minutes
30-paper batch Full marking day Minutes
Calibration to your standards Weeks of marker training 3 batches of corrections
Audit trail compilation Hours per module Seconds per export
Consistency across markers Varies marker-to-marker Deterministic (temperature = 0)
Availability Term-time, working hours 24/7
Onboarding a new marker Weeks One-time setup
Manual

~30 min

Marking Agent

<2 min

~15× faster per paper.
100%75%50%25%0%B1B2B3B4B5B6
Converges within 3 batches.

Verified cost-comparison and per-institution projected-savings figures coming as we close pilot data with our institutional partners.

Built for trust

Research-grade tooling, UK infrastructure.

Data residency
Azure UK South. Data does not leave the UK.
GDPR Article 28
Standard DPA template. Sub-processor list public.
Research-grade
Built and tested with UK university partners. Pilot / pre-commercial.
Calibration verified
Temperature = 0 across all calls. Every paper auditable.
Pilot data preserved
Read-only on lock — never deleted.
FAQ

Common questions.

Will the Agent grade exactly like I do?

Not on day one. The Agent grades against your mark scheme out of the box. As you correct its drafts, it calibrates to your standards. Most markers see meaningful convergence by their third batch.

What if my marking standards change mid-module?

You can reset the calibration on a per-mark-scheme basis. Or layer a new criterion-weighting profile without losing the historical record.

Is the Agent a black box?

No. Every score comes with criterion-level evidence, a written justification, and an audit trail of corrections. External examiners can read the rationale without translation.

Can I trust it on borderline cases?

The Agent surfaces low-confidence grades for human review before they ship. You see “Awaiting Calibration” or “Calibrating” status badges — explicit, not hidden.

What happens to my data if I stop using it?

Pilot data is preserved in read-only state when access ends — your work is not deleted. Sub-processor list public.

How can I get hands on it?

We’re working with UK university partners on pilots and R&D collaborations. Contact us to discuss testing, demo access, or a research partnership for your institution.

Media

Asset pack

Product deck, flyer, and 50-second explainer video.

Product deck
6-page PDF
Product flyer
1-page PDF
Explainer video
50 seconds

Ready to see it on your own marking?

We’re working with UK university partners on pilots and R&D collaborations. Contact us to scope what’s right for your institution.