Book A Consultation

Math Angel: personalised GCSE maths tutor at scale

Reduce support workload, guide buyers, and resolve complex queries. Our AI agents connect to your systems, ensure smooth handovers, and demonstrate ROI with clear analytics.

Math Angel product preview

Overview

Math Angel is a fast‑growing UK maths edtech platform. We connected an AI Tutor to user and content data, GCSE past papers, mark schemes, topic taxonomy, and difficulty tiers to deliver targeted guidance at scale.

Objectives

Improve personalisation and study outcomes, cut repetitive support, and gain clear analytics on topic mastery and content gaps.

Scope

AI Tutor (chat), mark‑scheme‑grounded hints, user‑level personalisation, handover with AI summaries, ticket dispatch, and analytics dashboards.

The Challenge

Deliver accurate, personalised help across thousands of questions while keeping knowledge consistent as content evolves.

Content consistency at scale

New content and updates risk conflicting guidance without governance.

Personalisation from real learner data

Guidance should adapt to past performance, conversation history, and quiz results.

Operational load

Repetitive queries tie up human support; handover should be precise when needed.

The Solution

A governed knowledge stack with precision retrieval and a great tutor experience that handover safely.

Data foundation

Ingested user profiles, conversation history, quiz results, platform resources, past papers, mark schemes, and topic/difficulty tags (Foundation, Additional Foundation, Higher Only).

KB Watchdog (knowledge quality)

Automatically flags contradictory or stale entries so guidance remains consistent over time.

Hybrid Graph + Vector retrieval

Graph captures relationships (topics, subtopics, difficulty); Vector search handles semantics. Together they boost precision and recall.

Tutor experience

Conversational help with source citations, step‑by‑step hints from mark schemes, and topic‑aware suggestions aligned to each learner's profile.

Handover & ticket dispatch

When confidence or permissions fall short, the agent creates a ticket with an AI summary and proposed resolution, and routes it to the right queue.

Analytics dashboards

Visibility into engagement, topic mastery, unanswered intents, and KB quality alerts, all filterable by cohort and time period.

Implementation

A structured, phased approach from discovery to ongoing optimisation.

Weeks 0-1

Discovery & ingestion

Connect data sources; map topics, difficulties, and mark‑scheme references; define guardrails.

Weeks 2-3

Tuning & QA

Retrieval accuracy checks, contradiction tests via KB Watchdog, and safety/policy tuning.

Week 4

Pilot launch

Limited cohort, analytics baseline, support playbooks for handover.

Ongoing

Optimisation

Quarterly KB health reviews, unresolved‑intent mining, and dashboard‑driven content improvements.

Outcomes

Clear learner and operational impact, with governance built‑in.

Student impact

Faster, cited answers with structured hints; nudges aligned to weak topics and recent errors; higher returning sessions.

Operational impact

Fewer repetitive tickets; cleaner dispatch; staff time shifts from FAQs to high‑value support.

Quantitative KPIs (for release)

Change in engagement, deflection rate, watchdog alerts resolved.

Why This Worked

Precision retrieval, quality control, and a data‑aware tutor loop.

Structured knowledge + taxonomy

Topic and difficulty mapping enable targeted hints and practice sequencing.

Mark‑scheme grounding

Hints mirror exam‑board reasoning with clear steps and definitions.

Safety & compliance

Least‑privilege access, audit logging, and child safeguarding.

Transferability to Your Use Case

The same stack generalises to other environments.

B2B SaaS: Ticket deflection

Deflect common queries with cited answers; keep edge cases for your team.

Learn more →

Shopify: Revenue & Support Copilot

Pre‑purchase guidance, wrong‑fit return prevention, and next‑best actions.

Learn more →

Industrial: Component Finder

Map spec‑to‑part accurately across dense catalogues and data sheets.

Learn more →

FAQ (about this deployment)

The tutor cites exam‑relevant steps correctly, so students understand the marking, not just answers.

Access is least‑privilege; PII is minimised; retention and audit are configurable per policy; ALL personal data fully anonymised.

KB Watchdog flags contradictions; editors review and resolve to maintain consistency.

Source‑first prompting, retrieval constraints, citation requirements, and confidence thresholds tied to “I don’t know” answers or handovers.

Yes. It can create tickets with summaries and proposed solutions, and other actions via approved integrations.

Ready to see this in your environment?

Get a walkthrough tailored to your stack and KPIs.

Book a Consultation