Webinar: AI Can Mark Text. But Can It Mark Video?

Learn more about AI Marking solutions by sAInaptic >

AI marking of written text is well established. Automated scoring engines have been processing essays, short answers and constructed responses for years, and the technology is now embedded in assessment workflows across the sector. But what happens when the evidence is not text on a page? What happens when an apprentice films themselves donning PPE in an engineering workshop, or a hospitality candidate is recorded managing a guest checkout at a hotel reception desk?

That was the question at the centre of a Ufi VocTech Trust-funded innovation project delivered by sAInaptic, working alongside two awarding organisations with very different assessment challenges: SIAS, a STEM-focused awarding body assessing practical engineering skills, and CTH Awards, the Confederation of Tourism and Hospitality, assessing customer-facing competence in hospitality settings. The project set out to discover whether AI marking technology could be extended beyond text to assess image and video-based evidence in real vocational contexts.

On 15 September 2026, the Test Community Network brought the project partners together for a live panel discussion to share what they found, what worked, and where the technology still has ground to cover.

On the panel

Jane Holmes, Associate Director of Grants, Ufi VocTech Trust

Deborah Hoggett, Head of Product and Assessment, SIAS

Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

Chaired by Tim Burnett, Test Community Network

Why this project exists

Apprenticeship completion rates in England sit just above 50%, with higher dropout rates at Level 3 and below. Of those who leave, 70% cite quality of training and lack of support as key reasons. Feedback on portfolio submissions, which form a significant proportion of end-point assessment evidence, is often delayed by four to six weeks. That delay makes it harder for apprentices to track which knowledge, skills and behaviours they have demonstrated, where gaps remain, and what to focus on next.

The project aimed to change that by delivering instant, personalised, examiner-level feedback on image and video portfolio submissions. By automatically mapping what the AI sees and hears in a submission against required competencies, the tool identifies demonstrated skills and highlights areas for development.

Jane Holmes explained why the project stood out from a competitive field of applicants to Ufi VocTech Trust's VocTech Activate programme.

"We really liked the innovative approach, answering the question: AI can mark text, but can it mark video? We all know that it can take tutors and trainers a long time to look through photo, video and multimedia submissions. We thought this solution could offer a new way to support vocational skills training. And that is what we are all about."
— Jane Holmes, Associate Director of Grants, Ufi VocTech Trust

She was equally clear that the strength of the team mattered as much as the idea itself.

"We also thought that the team at sAInaptic had a really good understanding of their learner audience and of the challenge that they were trying to address. Those were the two key things that sang out to us."
— Jane Holmes, Associate Director of Grants, Ufi VocTech Trust

Two very different assessment challenges

The project was deliberately designed around two contrasting use cases to test whether the approach could work across different assessment domains.

SIAS focused on formative learner support. Their interest was in whether AI could provide apprentices with structured feedback on practical evidence, specifically PPE compliance within engineering and manufacturing settings, to help them build stronger portfolios ahead of formal assessment. Deborah Hoggett described the thinking behind their involvement.

"What really interested us was whether AI could provide an additional source of feedback and guidance from image and video content that helps learners reflect on their performance and better understand the skills and behaviours they are expected to demonstrate."
— Deborah Hoggett, Head of Product and Assessment, SIAS

She was clear about the boundaries from the outset.

"The goal was not to replace assessors or to make assessment decisions. What we wanted to understand was whether the technology could help learners identify areas of development and improve their skills."
— Deborah Hoggett, Head of Product and Assessment, SIAS

CTH Awards came at it from a different angle entirely. Their moderators were reviewing recorded practical assessments of hospitality candidates: watching videos of reception scenarios, culinary practicals and professional skills demonstrations. The volumes were significant. Angela Hagenow, who has over 30 years of experience across hospitality operations, HR and education, described the scale of the challenge.

"Two thirds of our business is culinary practical and professional skills practicals. We would walk behind our colleagues' desks and they would be watching TV all day, but not TV -- maybe watching the same bed being remade twenty-five times. There had to be something better."
— Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

She explained that moderators were not simply watching and scoring. The workload extended well beyond the video itself.

"It is not just sitting through a thirty-minute video. It is also following that video: you have got to write up your notes, consider the assessor's decisions, follow up any concerns, re-look at some aspects, and then create feedback for the centres."
— Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

CTH's moderators are qualified professionals with deep experience in hospitality and tourism, and the goal was to free them from the mechanical burden of watching hours of footage so they could apply their expertise where it matters most.

How sAInaptic approached the technology

Dr Rajeshwari Iyer described an approach built on starting small and scaling deliberately. The team chose one specific qualification and one specific module within each awarding organisation, then broke the assessment challenge down into discrete components: beginning with object identification in images before progressing to video analysis, body language, softer skills and more complex behavioural assessment.

"We set ourselves very tangible, realistic goals. We started with image assessments first, because if you think about what videos are, they are basically a collection of lots of different images. We were not trying to do everything from the beginning. This project has been ongoing for a year, and we have had engineers working on this full-time for nearly eight months."
— Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

A critical technical distinction underpins sAInaptic's approach. Their model is a hybrid, combining computer vision with their existing marking engine rather than simply bolting generative AI onto a large language model. This matters for reliability. In assessment, consistency and explainability are not optional extras. They are fundamental requirements.

"Because our model is a hybrid model, there is no hallucination. Repeated runs on the same assessment showed less than one mark deviation. It is highly repeatable and highly reliable."
— Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

The project data backs this up. Across the CTH video assessments, the AI achieved a mean absolute error of 1.6 marks, well within inter-rater variability (which sits at 5 marks out of 25 and 6.5 marks out of 50 for human assessors). Repeated runs on the same video produced less than one mark deviation, demonstrating a level of consistency that human marking panels typically struggle to match. When the vision model was trained on video annotations, it performed remarkably well. And once the model was fine-tuned with positive reinforcement data, the accuracy on that dimension was described as flawless.

But Dr Rajeshwari Iyer was equally candid about where the technology struggled, and this honesty is itself a mark of credibility in a space where overpromising is common. The vision model cannot see what it has not been told to see.

"AI does not know what it does not know. We did not do very detailed video-based annotations, so hand movements, gestures, body language -- the AI struggled to repeatedly and reliably identify, for example, cheating, which assessors from having done this for years can spot from certain eye-to-hand movements."
— Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

That kind of openness about limitations, combined with the rigour of the underlying approach, is what separates a serious assessment technology from a demo built on top of a chatbot.

What the feedback looks like in practice

For SIAS, the AI produces structured feedback showing what the apprentice said they were wearing, what the AI actually identified in the video, and how that maps against workplace standards and the relevant knowledge, skills and behaviour statements. The feedback tells the apprentice clearly which items were present, which were missing, and how compliant they were overall. It then maps what it observed against the specific KSB statements relevant to that qualification module.

For CTH Awards, the output mirrors the structure that human assessors already use, providing narrative feedback and dimensional scores against the five main assessment criteria: whether the student met the assessment criteria, technical skills and knowledge, communication skills, personal presentation, and self-confidence and effort.

Angela noted the quality had exceeded expectations.

"The feedback is quite detailed and far more extensive than we thought we would have. And we have noticed that with each iteration, each time we have given feedback to Raj, things have improved."
— Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

One of her colleagues, initially sceptical about the project, had come around entirely. As Angela put it during the session: "Yesterday one of my colleagues said, 'I am amazed at its ability to do the job.'" For Angela, the value was clear: the tool was letting her team do what they were hired to do.

"I want to see our experienced moderators using their academic skills and their expertise where it counts rather than just slogging through so many videos."
— Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

The feedback loop between the awarding organisations' subject matter experts and sAInaptic's engineering team proved essential throughout. Dr Rajeshwari Iyer gave a telling example of how the quality of training data shapes the quality of AI output.

"Within the feedback that we originally saw from assessors, there was not that much positive reinforcement. So that is exactly how the AI was giving its own feedback. But once Melissa, who was working with us, added the positive reinforcement, the final fine-tuned model now does as much positive reinforcement as it was doing negative. It is really just working with assessors and tutors to come up with that gold standard data set that can then be used to train the model."
— Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

The AI reflects the quality of what it is trained on. That is precisely why domain expertise and close collaboration with assessment professionals is not optional. It is foundational.

Expert in the loop: augmenting, not replacing

The panel were straightforward about the regulatory context. Ofqual's current position prohibits the sole use of AI in marking, and the panel were clear that these tools are designed to augment the assessment process rather than replace the professional judgment that sits at its centre. Deborah Hoggett framed this not as a limitation but as a deliberate and responsible choice.

"We are an awarding organisation, we operate within a regulated environment. High-stakes assessment decisions need to think about validity, fairness, trust. All those principles remain central to everything we do."
— Deborah Hoggett, Head of Product and Assessment, SIAS

But she was clear that the project had demonstrated real value within those boundaries.

"What has been really exciting is that we have worked with Raj and the team at sAInaptic and we have demonstrated that AI can generate useful feedback on practical evidence. We just need to then consider how the technology in the future can complement that human expertise."
— Deborah Hoggett, Head of Product and Assessment, SIAS

Angela echoed this from the moderation perspective.

"We would never consider it being the only moderator. But it is so much easier to moderate an even smaller sample than we moderate now."
— Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality (CTH)

The underlying principle is clear: the expert remains in the loop. Effort in the loop still matters too. The quality of the AI output is directly tied to the effort that goes into training it, refining it and working alongside it. That is not a weakness of the technology. It is a feature of doing it properly.

What comes next

sAInaptic is targeting commercial availability by the end of Q1 2027. The immediate technical challenge is proving that the model generalises across qualifications and assessment types. Within SIAS alone, moving from PPE compliance to other practical demonstrations within the same qualification is the next step. CTH Awards has culinary qualifications with entirely different assessment scenarios to explore. The scope for expansion within the existing partnerships is significant before the technology is taken to the wider market.

For other assessment technology providers, sAInaptic's approach also opens the door to white-label integration: embedding AI video and image marking capability within existing platforms and workflows rather than requiring organisations to adopt an entirely new system.

Demand signals are encouraging. As Dr Rajeshwari Iyer noted, awarding organisations and training providers are increasingly moving towards video-based assessments as a more authentic form of evidence, particularly given the growing use of generative AI by learners to produce their submissions. That shift could accelerate demand for exactly this kind of marking capability.

"More recently, awarding organisations and other centres are moving towards video assessments as a more authentic way to do their assessments, given how much AI is being used by learners to submit their assessments. So there is a need."
— Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic

Ufi VocTech Trust: more than the money

This project was funded through Ufi VocTech Trust's VocTech Activate programme, an annual innovation grant call providing up to £60,000 over twelve months for projects that develop technology to support people in gaining skills for work. The Trust has worked with over 250 organisations across more than 300 projects to date, and their next VocTech Activate call opens on 5 January 2027.

What sets Ufi VocTech Trust apart is the support that surrounds the funding. This is an organisation that exists to back innovation in vocational learning, and they go well beyond writing a cheque. Their focus is the learner, and everything they do is designed to give projects the best chance of creating real, lasting impact for the people gaining skills for work.

"We are not like other funders. We take a 'have you thought about' approach, because sometimes we can see opportunities or clouds on the horizon, or maybe a bear trap that they have not seen. We never tell people what to do, but we do try to help them understand where the path might go."
— Jane Holmes, Associate Director of Grants, Ufi VocTech Trust

That support includes free guidance on intellectual property protection, commercial positioning, pricing strategy, and identifying wider market opportunities. It is a model designed to help innovative ideas become self-sustaining products that reach as many learners as possible.

She emphasised one factor above all others.

"Any technology developed in isolation of its user group is rarely a success. Having the connections with the people who will be using this new technology, understanding what each of them wants from it right from the very start and not halfway through, is an absolutely critical part of these developments."
— Jane Holmes, Associate Director of Grants, Ufi VocTech Trust

For anyone with an idea they think Ufi VocTech Trust might back, sign up for the newsletter at ufi.co.uk. Pre-application webinars will be announced ahead of the January call, giving prospective applicants the opportunity to discuss their ideas directly with the grants team before submitting.

The question that matters

The webinar set out to answer a simple question: can AI mark video? The evidence from this project suggests that the answer is yes -- within defined contexts, with the right training data, with domain experts closely involved, and with a technology partner whose approach prioritises reliability and explainability over speed to market. It is not a finished product. It is a proof of concept that has demonstrated something genuinely new, and the organisations involved are already planning what comes next.

For assessment leaders watching from other awarding organisations, the question is no longer whether this is technically possible. It is whether your organisation is ready to explore it.

This article is based on a live panel discussion hosted by the Test Community Network on 15 September 2026. The full webinar recording is available on demand at testcommunity.network.

Guest Profiles

Deborah Hoggett, Head of Product and Assessment, SIAS
Leading the development of industry-focused qualifications and apprenticeship assessments across the STEM sectors. Deborah works collaboratively with employers and industry experts to create robust, flexible, and relevant qualification and assessment solutions that meet the changing needs of learners and industry. With a particular interest in innovation, Deborah champions new approaches to qualification and assessment design, learner support, and assessment readiness, helping organisations provide effective guidance and preparation that enables learners to succeed. Deborah is passionate about sharing best practice and supporting organisations to navigate changing skills demands while maintaining quality and compliance.

Jane Holmes, Associate Director of Grants, Ufi VocTech Trust
Jane is responsible for leading and managing Ufi's grant funding work, with overall responsibility for projects and cohorts from call scoping through to onboarding, monitoring, forecasting, delivery, enrichment, closure, and evaluation. Her career spans grant programme management across organisations including Birmingham Science City, Innovate UK, and the Worcestershire Proof of Concept Fund, as well as innovation strategy work with the West Midlands Combined Authority. A background in global sales negotiation brings a strong customer and user focus to Ufi's work. For this webinar, Jane brings the funder's perspective on the project that asked whether AI marking technology could be extended to assess video-based evidence in vocational settings.

Angela Hagenow, Academic Director, Confederation of Tourism and Hospitality - CTH
Angela started her career in hospitality and tourism operations before moving into HR; she has over 30 years’ experience in all aspects of education, employment and people development. After studying hospitality management Angela worked in food and beverage departments in Hilton Hotels in Munich and London, and then had personnel and training management positions within Intercontinental Hotels and Marriott Hotels before moving on to teach business management in the university sector for eight years. Angela has also worked as a training & development consultant for private industry, commerce, professional firms and the public sector. Initially as Chief Examiner and now as Academic Director of CTH, the Confederation of Tourism and Hospitality, Angela is responsible for managing the academic function, organisation and development of CTH qualifications from Level 3 to Level 7 in hospitality and tourism. Some of her responsibilities include developing the qualifications and assessments, monitoring and analysing student performance data, and using that data to strengthen CTH’s academic programme. Angela is a member of CTH’s Accreditation Panel, Curriculum Development Team, Examination Board and Academic Council and has represented CTH at conferences and seminars. Part of her role is regulatory compliance and coordinating academic and policy dissemination; she has set, marked and moderated examinations and assignments and is dedicated to helping students reach their full potential.

Dr Rajeshwari Iyer, Co-Founder and CEO, sAInaptic
Rajeshwari holds a PhD in computational neuroscience from Imperial College London and an MPhil from the University of Cambridge, with over 12 years of experience in computational modelling and data science. As Co-Founder and CEO, she leads sAInaptic's mission to transform assessment through intelligent automarking. For this webinar, Rajeshwari brings the technology perspective, sharing how sAInaptic's AI marking model was extended to assess video-based evidence as part of the Ufi VocTech Trust-funded project.