Policy and AI governance
AVA is built to be defensible in a risk assessment, a data protection impact assessment and a board paper. We align with UK education and workforce policy, lead with safeguarding, and keep our evidence language honest.
How we align
UK guidance for generative AI in education converges on four expectations: filter content, monitor use, protect data, and govern decisions transparently. AVA is engineered against all four from the first student input onward.
Governance, in place
Each governance expectation is built in today, stated as a plain binary. Full documentation is shared during institutional due diligence.
The standards we build to
These are the public standards an institution will check AVA against. We treat them as the floor, not the ceiling.
AVA is built to the Department for Education expectations for generative AI in education: age-appropriate content filtering, monitoring of prompts and responses, strong data protection, child-centred design and clear governance.
For an audience that is mostly 16 to 18, we apply age-appropriate design, data minimisation and high-privacy defaults, in line with the Information Commissioner Children’s Code.
Safety-by-design and documented risk assessment for content that reaches children, consistent with Ofcom protection-of-children duties.
Built to the Department for Education expectations for generative AI in education: filtering, monitoring, strong data protection, child-centred design and clear governance.
Read the standardAge-appropriate design, data minimisation and high-privacy defaults for an audience that is mostly 16 to 18.
Read the standardSafety-by-design and risk assessment for content that reaches children, in line with Ofcom protection-of-children duties.
Read the standardTriple-lock safeguarding on every student input path, with clear escalation to designated safeguarding leads.
Read the standardDesigned for scrutiny
Documentation, escalation routes and evidence gates are part of the product, built and in place before a partner asks for them.
Safeguarding
Students are 16 to 18, so safeguarding is designed in from the first input onward. Every input path is moderated, age-appropriate, logged and protected, with clear escalation to the people responsible.
Each student input passes through layered moderation before any AI processing, with a curated term list for the medical-education context students work in.
Serious concerns suspend the session and escalate to designated safeguarding leads, in line with Keeping Children Safe in Education.
Access is paused overnight using the device clock alone, so no location or timezone data is collected, protecting students seeking support outside appropriate channels.
Inputs and responses are logged so a partner can audit decisions, review flags and tune sensitivity over time rather than trusting a black box.
AI and access
Who reaches good AI already tracks income and education. The families who could always buy a tutor are the ones buying an AI tutor, while a student with no professional network gets whatever is free, ungoverned and built for someone else. AI is fairer than the access it replaces only where it is built that way on purpose: governed for the age group using it, declared on every surface, and put in the hands of the students who have the least.
Students practise against a system built for their age and their task, inside guardrails, rather than against a general assistant that was designed for neither.
Each part of the platform states plainly whether AI is involved and what it is doing there, so a student always knows what they are talking to.
The habit AVA builds is verification: where a claim came from, what it rests on and how to test it. That habit is the difference between using AI well and being used by it.
Every task has a path that works without the AI layer, so a student who prefers to think it through on paper reaches the same place.
Workforce and access policy
Policy alignment reaches well past safety. AVA is built to support the national agenda to widen access to selective courses and to grow a broader, fairer professional and clinical workforce.
Supports the expansion of medical and clinical training places by preparing a broader, fairer pool of applicants for selection.
Read the standardDesigned to sit inside universities access and participation commitments and contextual-admissions frameworks.
Read the standardAligned with national social-mobility goals: helping students without private networks reach selective courses and careers.
The national programme trains 10 million workers in AI skills by 2030, starting from the point people are already in work. AVA builds the same foundations earlier, while a young person is still deciding what work to aim at.
Read the standardThe national tutoring programme reaches up to 450,000 disadvantaged pupils in Years 9 to 11. AVA picks the same students up at the post-16 selective stage, where the gap in offers is decided.
Read the standard3
national safety standards aligned to
UK
hosted and data-sovereign by design
DPIA
and consent workflows built into scoping
AVA students are 4x more likely to get into medical school (92% versus 22%), and 2x more likely to get into Oxbridge across subjects.
The cited landscape
Our alignment with national standards answers something concrete: documented gaps in attainment, in access to the professions and in the medical workforce, each drawn from public sources.
1.01m6
16 to 24 year-olds are not in education, employment or training, the first time the total has passed one million since 2013.
Explore£125bn7
the estimated cost to the UK each year of nearly one million young people not in education, employment or training.
Explore19.1 months8
the gap in learning between disadvantaged pupils and their peers by the end of secondary school, wider than before the pandemic.
Explore2x9
disadvantaged young people are twice as likely to be out of education, employment or training as their better-off peers.
Explore7x1
more likely to reach Oxbridge if you attend an independent school rather than a state school.
Explore36% vs 6%2
of senior professionals were privately educated, against six per cent of the population.
Explore5%13
of medical school entrants are from working-class backgrounds, against 75 per cent from the most advantaged group.
Explore1.5x13
more likely to receive a medicine offer from an independent school, even after adjusting for grades and background.
Explore6% to 14%12
the rise in UK medical entrants from the most deprived areas over the decade to 2023: real progress, not yet parity.
Explore25,7703
applicants competed for around 8,100 UK medical school places in the 2026 cycle.
Explore100,02014
full-time-equivalent vacancies across NHS England, a vacancy rate of 6.7 per cent.
Explore260k to 360k5
the projected NHS staff shortfall by 2036/37 without sustained action on training and retention.
ExploreReversed15
the 2023 pledge to double medical school places to 15,000 is now being wound back, even as competition for places rises.
ExploreCold spots16
opportunity for young people is sharply unequal by place, concentrated in the North East, North West and former industrial areas.
Explore21.9%8
of disadvantaged students had no substantial education activity at the start of Year 12, against 9.3 per cent of their peers.
Explore£19bn17
in GDP the UK forgoes each year because talent from lower socio-economic backgrounds does not reach the roles it could fill.
Explore10m18
workers are to be trained in AI skills by 2030 under the national programme, which starts at the point people are already in work.
Explore450,00019
disadvantaged pupils in Years 9 to 11 are the target of the national AI tutoring programme, which reaches them before the selective stage decides who gets through.
ExploreSources
Anatomy of the gap
When access to selective routes is left to chance, the people who get through look very different from the population. These are public figures, each carrying the source it came from.
Privately educated, by profession10
Senior judges
Finance directors
Members of Parliament
The population overall
2009 snapshot from the Panel on Fair Access to the Professions, against 7 per cent of the population in independent schools.
Competition for a medicine place3
UCAS applicants to medicine against the places available, 2026 cycle.
Medical entrants, by background13
Most advantaged group
Lowest socio-economic group
Share of medical school entrants by socio-economic group, Unequal Treatment, 2025.
The arc
The case for fair access has been made and remade since the Milburn review named the professions a closed shop. The gap has narrowed, but it has not closed. AVA is built to be the next, measurable step, and to prove it honestly.
2009
The Milburn panel found medicine and law the most socially exclusive professions, with doctors typically growing up in the best-off families of any profession.
2014
UK medical schools commit to contextual admissions, outreach and alternative routes to widen who applies and who is selected.
2019
The country's leading people remain five times more likely to have been privately educated than the population as a whole.
2025
The independent-school share of medicine entrants has fallen, yet just 5 per cent of entrants are working-class and advantaged applicants are still around 1.5 times more likely to get an offer.
2026
Youth disengagement passes one million for the first time since 2013, at an estimated cost of £125 billion a year.
Evidence review
Evaluation work tests whether structured preparation changes offer rates. Causal language requires comparator definitions, documented limits and formal review.
What we can show
Beyond the public picture, partners get internal cohort analysis, comparator definitions and the full governance documentation during due diligence.
Evidence discipline
Governance covers honesty as much as safety. Each figure is labelled with the kind of evidence it is: engagement, observed in cohort, or measured against a comparator. Observational cohort findings and causal claims are labelled separately, and read separately.
Type 1
Practice sessions, time on task and completion. Reported as delivery monitoring, and labelled as engagement wherever it appears.
Type 2
Internal analysis of how prepared cohorts moved through interview and offer stages against relevant baselines. Honest, defensible, and clearly observational rather than causal.
Type 3 (formal review)
Effect estimates that rely on comparator definitions, documented limits and formal review before being used as causal evidence.
What we report
Causal language requires comparator definitions, documented limits and formal review. Every figure travels with the evidence type behind it, so a reader always knows which kind they are looking at.
Evaluation design
The allocation, the comparator and the primary outcome were all fixed before the evaluation opened, so the finding reads the same whichever way it lands. Five medical school research partners share admissions outcome data under formal agreements, which is what lets an offer rate be measured rather than estimated.
How the evaluation is designed
Design, comparator and primary outcome agreed in advance. Findings follow the outcome window and formal review, and carry their limits with them.
What we can show
Bring AVA to your evaluation team, your data protection lead and your access board. This is the level of detail we can put in front of them during due diligence.
Internal cohort analysis available for institutional review.
Evaluation design and comparator definitions available during due diligence.
GDPR, DPIA, consent and data-sharing workflows built into partnership scoping.
Outcome definitions and comparator logic documented in partner reporting.
From pilot to infrastructure
Evidence is a by-product of a disciplined partnership, not a report bolted on at the end. The same governed sequence runs behind every engagement.
We map the cohort, admissions route, current provision, staff workflow, data permissions and the outcomes that matter, before recommending anything.
A pilot scoped around a defined cohort and clear success measures, with baselines agreed up front so it can be evaluated honestly.
Measurement, outcome linkage and reporting that an evaluator can defend, with governance and safeguarding documented for risk assessment.
Evidence gates at pilot, renewal and expansion turn a successful pilot into an embedded, governable part of strategy, not another disconnected tool.
Data protection in practice
For an audience that is mostly 16 to 18, data protection is designed in. Here is how AVA answers the questions that come up in a procurement form and a data protection impact assessment.
We collect only what a student needs to practise and progress. No location or timezone data is collected, and overnight access pauses use the device clock alone.
Student data is hosted in the United Kingdom by design, so a partner can answer the data-residency question in a procurement form without a caveat.
Consent and information workflows are built into scoping, with age-appropriate language for a 16 to 18 audience and the information a school or guardian expects.
Retention periods are agreed with the partner, and deletion and export routes are defined up front rather than negotiated after a request arrives.
Student data is not sold and not used to train third-party models. It is used to run the service the partner has commissioned, and nothing else.
Access, correction and erasure requests are supported through a defined process, with a named contact and a documented response route.
How a review runs
A governance review runs with your team, step by step, from scoping through to the cadence that keeps it live long after launch.
We help map the cohort, the data flows and the lawful basis, and support the data protection impact assessment from the start rather than at sign-off.
Filtering, monitoring, safeguarding and data handling are written up in a pack an evaluation team, a data protection lead and an access board can each read.
We walk the escalation route end to end, from a flagged input to a designated safeguarding lead, so the people responsible see exactly how it behaves.
Sensitivity, logging and reporting are reviewed on a cadence the partner sets, so governance stays live rather than a one-off form at procurement.
Evidence guardrails
Governance is about clarity as much as safety. Here is every kind of claim AVA makes, the label it carries and the thing that backs it.
AVA is built to UK education and workforce policy and is defensible in a risk assessment, a DPIA and a board paper.
Backed by the full documentation pack, shared during institutional due diligence.
Every student input is moderated and logged, and safeguarding escalation runs to designated leads.
Backed by an end-to-end walkthrough, from a flagged input to the named lead who receives it.
Observational cohort findings are held now, presented with their comparators and caveats.
Labelled Type 2, observed in cohort, with the baseline and the limits alongside the figure.
Engagement and usage are reported as engagement and usage.
Labelled Type 1, and read as delivery monitoring in partner reporting.
An offer-rate claim carries the comparator definitions, the documented limits and the formal review that stand behind it.
Labelled Type 3, and used as causal evidence once that review standard is met.
Every figure travels with the cohort, the comparator and the outcome field it came from.
Backed by definitions an evaluation team can interrogate line by line.
Common questions
Explore more
Institutional partnership briefing
We will map your cohort, current provision, delivery constraints and evidence goals, then recommend a pilot or partnership shape that can stand up to institutional scrutiny.
Tell us where your foundations stand and we will talk through whether the AI-readiness work is a fit. Pick a time below, or email us if you prefer.
Prefer email? Write to contact@theaspiringmedics.co.uk, or open the booking page in a new tab.
