SUBTXT · Hackathon · 2026

Using AI to make expert video borderless

In a 24-hour hackathon, I teamed up with two developers to design and ship SUBTXT — a product that identifies and explains the local references in video conversations that global audiences miss.
expressed through a system blueprint and exploratory interface designs.

ACHIEVEMENTS

  • Identified a real-world use case for an emerging AI capability
  • Shaped how the AI decides what to  explain flag, skip, and explain
  • Made the product legible through brand and interface
  • Shipped the prototype in 24 hours, with two developers

Role

Strategy, Design, Branding

Team

Me, Developer, Developer

Me, Founders (CTO/PM)

Year

2026

ACHIEVEMENTS

Hackathon brief

Use Hera's API and DeepMind to build a creative agent that:

  • Solves a specific, real world problem
  • Makes editorial decisions autonomously
  • Operates according to defined beliefs and presets

3 research artifacts available on desktop

🔍 What Focus Mode's design revealed

Focus Mode's design suggested a narrow framing of the problem - treating distraciton as something external, to be managed by controlling the environment.

🧠 A behavioural lens

To better understand focus and distraction, I turned to behavioural frameworks. In Indistractable, Nir Eyal writes:

While we love to blame external triggers...most of our distractions begin from within

This reinforced my suspicion that we were conceiving the problem too narrowly and provided inspiration for concept testing.

📈 The commercial opportunity

This misalignment wasn't just theoretical, it was reflected in the tools available.

Heading goes here

Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Combining Hera and Gemini

Experimenting with Hera's API, we learned it was capable of surprisingly precise control over text and motion. Our idea was to pair it with DeepMind's Gemini model, which can analyse video directly, building a pipeline that could automatically caption video for content creators.

3 research artifacts available on desktop

🔍 What Focus Mode's design revealed

Focus Mode's design suggested a narrow framing of the problem - treating distraciton as something external, to be managed by controlling the environment.

🧠 A behavioural lens

To better understand focus and distraction, I turned to behavioural frameworks. In Indistractable, Nir Eyal writes:

While we love to blame external triggers...most of our distractions begin from within

This reinforced my suspicion that we were conceiving the problem too narrowly and provided inspiration for concept testing.

📈 The commercial opportunity

This misalignment wasn't just theoretical, it was reflected in the tools available.

Reviewing the Product

Hera’s API made it possible to generate controllable animated overlays with transparent backgrounds.

Contextual overlays are increasingly being adopted by content creators.

The real world opportunity

Looking for a real-world problem to solve with our technology, I pitched the team an idea: help video producers make content more accessible to international audiences by captioning local references they'd otherwise miss. It resonated, and we ran with it.

  • Users had already expressed interest in accountability-based support
  • It addressed a clear gap in existing focus tools
  • It aligned with behavioural research on how focus actually breaks down
  • Early thinking suggested the technical implementation was feasible

Slides from our submission deck, explaining the concept to judges.

Giving the agent a point of view

The backend was led by Sami, our developer. I wrote the prompts that decided what the agent noticed and how it looked: one instructing Gemini on what qualifies as a gap worth explaining, and one instructing Hera on how each caption should look and move on screen.

  • Users had already expressed interest in accountability-based support
  • It addressed a clear gap in existing focus tools
  • It aligned with behavioural research on how focus actually breaks down
  • Early thinking suggested the technical implementation was feasible

The pipeline combined AI reasoning with generative video — the reasoning steps make editorial decisions, the rest is production.

The pipeline
The Prompt
Tab 3

Designing the creator experience

The interface had three jobs: make an unfamiliar AI workflow feel self-explanatory, give creators basic control over the output, and give the product an identity people would remember. Alongside the interface, I designed the brand, naming and messaging.

  • Users had already expressed interest in accountability-based support
  • It addressed a clear gap in existing focus tools
  • It aligned with behavioural research on how focus actually breaks down
  • Early thinking suggested the technical implementation was feasible

The interface allows users to guide the agent regarding what gets captioned as well as caption density.

PROCESS 2/3

Example output

We handed off the livesite, readme. Example video below.

What worked well / Successes

  • Identifying a commercially plausible editorial workflow for agentic video
  • Making it legible to an unfamiliar target audience
  • Collaboratig effectively with developers to make the most of our talents
  • Shipping in 24 hours
⚠️ What worked well

Signal quality
‍Even when distraction is present, prompts may feel disruptive or punishing.

Privacy
‍Monitoring application and window behaviour may feel invasive to some users.

🗺️ Areas for further exploration
  • Simulate the experience
    Test focus prompts manually to understand whether they support focus or create friction, and for which users and contexts they are effective
  • Build a narrow prototype
    Validate drift detection on a single task
  • Expand to a full system
    Scale into more complex takss and introduce additional layers of support
  • Users had already expressed interest in accountability-based support
  • It addressed a clear gap in existing focus tools
  • It aligned with behavioural research on how focus actually breaks down
  • Early thinking suggested the technical implementation was feasible

Reviewing the Product

A video generated with SUBTXT using the biggest news story of the day.

Outcome and reflections

We weren't able to get feedback on the submissions, but the idea resonated enough with the team that we've talked about picking it back up where the priorities my side would be:

What worked well / Successes

  • Identifying a commercially plausible editorial workflow for agentic video
  • Making it legible to an unfamiliar target audience
  • Collaboratig effectively with developers to make the most of our talents
  • Shipping in 24 hours
  • Refining the overlay system’s motion language, timing and visual hierarchy
  • Testing more advanced customisation settings
  • Exploring use cases beyond political media and adapting value proposition and functionality accordingly
⚠️ What worked well

Signal quality
‍Even when distraction is present, prompts may feel disruptive or punishing.

Privacy
‍Monitoring application and window behaviour may feel invasive to some users.

🗺️ Areas for further exploration
  • Simulate the experience
    Test focus prompts manually to understand whether they support focus or create friction, and for which users and contexts they are effective
  • Build a narrow prototype
    Validate drift detection on a single task
  • Expand to a full system
    Scale into more complex takss and introduce additional layers of support
  • Users had already expressed interest in accountability-based support
  • It addressed a clear gap in existing focus tools
  • It aligned with behavioural research on how focus actually breaks down
  • Early thinking suggested the technical implementation was feasible

PROCESS 2/3

Handoff

Due to wider company priorities, the team chose to pause product discovery and launch the existing product as an open beta. I handed off a system blueprint, feasibility assessment, and validation roadmap — enabling the team to return to this direction when ready.

Feasibility & Risks

Signal quality (low–moderate risk)
‍Even when distraction is present, prompts may feel disruptive or punishing.

Privacy (moderate risk)
‍Monitoring application and window focus may feel invasive to some users. Clear opt-in, transparency, and local processing would be essential.

Proposed next steps
  • Manually simulate focus prompts
    to observe whether they support focus or create stress
  • Build a narrow prototype
    focused on a single, simple task to test whether attention drift can be detected reliably
  • Expand into more complex tasks
    once the core concept proves viable

The 6 Focus Concepts

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Smart Blocking

Apps and websites that are not helpful to your task are blocked.

Accountability

Background Monitor

See how closely your behaviour matches your intended task (always-on).

Accountability

Visual Environment Change

Apps and websites that are not helpful to your task are desaturated or blurred.

Accountability

Session Reports

Analytics show how you've spent your focus time and offer goals and tips.

Accountability

Timeboxed Focus

Plan focus sessions in advance. Smart reorganisation adapts to conflicts.

Accountability

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Smart Blocking

Apps and websites that are not helpful to your task are blocked.

Accountability

Background Monitor

See how closely your behaviour matches your intended task (always-on).

Accountability

Visual Environment Change

Apps and websites that are not helpful to your task are desaturated or blurred.

Accountability

Session Reports

Analytics show how you've spent your focus time and offer goals and tips.

Accountability

Timeboxed Focus

Plan focus sessions in advance. Smart reorganisation adapts to conflicts.

Accountability

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Accountability

Focus Environments

Self-awareness

Scheduling

Accountability

Focus Environments

Self-awareness

Scheduling

Smart Blocking

Visual Environment Change

Background Monitor

Accountability Prompts

Timeboxed Focus

Session Reports

Focus-Drift Prompts

Beams recognises when attention drifts from intended task and nudges you back on track.

Visual Environment Change

Apps and websites that are not helpful to your task are desaturated or blurred

Session Reports

Analytics show how you've spent your focus time and offer goals and tips.

Smart Blocking

Apps and websites that are not helpful to your task are blocked.

Timeboxed Focus

Plan focus sessions in advance. Smart reorganisation adapts to conflicts.

Background Monitor

Widget shows how closely your behaviour matches intended task.

Key learnings

Accountability emerged as the best opportunity for differentiation.

Focus Environments were seen as powerful but restrictive.

Self-awareness generated limited engagement.

Scheduling was a baseline expectation.

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Focus-Drift Prompts

Beams recognises when your attention drifts from your intended task and nudges you back on track.

Accountability

Handoff

I handed off the system blueprint, feasibility analysis, and validation roadmap - equipping the team with everything needed to move forward.

🧠  How attention drift is detected

When a user starts a focus session, they select a work profile that defines expected behaviour. If activity consistently falls outside the expected pattern, it flags this as attention drift and triggers a notification. Over time, the model adapts, learning what focused work looks like for each user and task type.

⚖️  Feasibility & Risks

Signal quality (low–moderate risk)
‍The system depends on reliably detecting attention drift from application and window focus patterns. Even when technically accurate, prompts that appear at the wrong moment may feel disruptive or punishing.

User acceptance / privacy (moderate risk)
‍Monitoring application and window focus may feel invasive to some users. Clear opt-in, transparency, and local processing would be essential.

🧭  Proposed next steps

  • Manually simulate focus prompts to observe whether they support focus or create stress
  • Build a narrow prototype focused on a single, simple task to test whether attention drift can be detected reliably
  • Expand into more complex tasks once the core concept proves viable

Awareness Moments

Personal Accountability

Beams recognises when your attention drifts from your intended task and nudges you back on track

This would be a game-changer if it worked correctly

I assumed this wasn’t possible!

It might be a fine line between helpful and annoying...

Awareness Moments

Personal Accountability

Beams recognises when your attention drifts from your intended task and nudges you back on track

Awareness Moments
Smart Blocking
Background Monitoring
Visual Environment

Awareness Moments

Personal Accountability

Beams recognises when your attention drifts from your intended task and nudges you back on track

QUOTES

This would be a game-changer if it worked correctly

It might be a fine line between very helpful and annoying

RATINGS

Usefulness

Usefulness

Usefulness

PROCESS 1/3

My Process

DISCOVERY

Reframing the problem

Beams' goal was to help users focus and reduce context switching. But despite positive feedback during the closed beta, its core feature — Focus Mode — saw less than 10% repeat usage.
‍
To the team, this looked like a usability issue. I suspected it was more fundamental. The product treated focus as something that could be set once upfront and maintained by controlling the environment - without accounting for how attention can drift during work. Rather than refining this approach, I proposed exploring alternative paths forward.

What Focus Mode's design revealed

Focus Mode's design suggested a narrow framing of the problem - treating distraciton as something external, to be managed by controlling the environment.

On the surface, focus mode covered the problem space. In practice it relied on a flawed assumption.

Understanding the problem through a behavioural lens

To better understand focus and distraction, I turned to behavioural frameworks. In Indistractable, Nir Eyal writes:

"While we love to blame external triggers...most of our distractions begin from within"

This reinforced my suspicion that we were conceiving the problem too narrowly and provided inspiration for concept testing.

The commercial opportunity

This misalignment wasn't just theoretical, it was reflected in the tools available.

3 research artifacts available on desktop

Analysing the beta product

→

Behavioural framework

→

Competitor Landscape

→

Competitor Benchmarking
Expert Research
Focus Mode 1.0 Analysis

Benchmarking revealed which focus solution areas were saturated and which were underserved.

"

Most distraction doesn’t originate from outside of us... most distraction starts from within.

Nir Eyal, Indistractable

Established frameworks helped dmonstrate why we should broaden our understanding of the problem

Focus Mode 1.0 attempted this by combining timers, intention setting and Do Not Disturb functionality.

Established frameworks helped demonstrate why we should broaden our framing of the problem

Benchmarking revealed which focus solution areas were saturated and which were underserved.

Heading goes here

Lorem ipsum dolor sit amet, consectetur adipiscing elit.

+

Focus Mode 1.0 analysis →

+

Expert research →

+

Competitor Benchmarking →