See
Vision Navigation
Real-time object detection using YOLOv8 gives audio cues for obstacles, signage, and surroundings — turning any phone camera into a navigation companion.
Object Detection · Audio Feedback
A background illustration of three circles representing the barriers EBITH removes: a blurred street scene fading behind a closing vignette, a sound wave that repeatedly flatlines into silence, and a sentence that freezes mid-word above a mouth held open. As you scroll, the three circles merge into a single circle glowing teal with an unbroken wave running through it.
AI-Driven Accessibility
Even Blind,
I Am The Hero.
EBITH gives visually impaired and deaf individuals real-time AI tools for navigation, communication, and independence — built for humans, not demos.
93 BSL Signs Recognised92.4% Accuracy
By The Numbers
The scale of the problem, in figures.
- Deaf people in the UK
- 2M+
Deaf people in the UK
RNID, 2024
- Registered blind and partially sighted people in the UK
- 340K+
Registered blind and partially sighted people in the UK
RNIB, 2024
- UK assistive tech market by 2030
- £4.2B
UK assistive tech market by 2030
Projected CAGR 7.4%
- Our BSL recognition accuracy
- 92.4%
Our BSL recognition accuracy
Held-out test set, 93 signs
What EBITH Does
Three barriers. One app. No hardware to buy.
Sign
BSL Recognition
93 British Sign Language signs recognised at 92.4% accuracy using MediaPipe landmarks and a custom-trained GRU model — running entirely on-device.
On-Device ML · No Cloud Required
Speak
Communication Bridge
Two-way communication: sign language translates to speech for hearing people; speech converts to large-display text for deaf users. No interpreter needed.
Real-Time · Bidirectional
Inference Pipeline
Camera Input
30 fps video stream · no upload
MediaPipe Hands
21 landmarks × 3 axes = 63 features
GRU Sequence Model
30 frames · 258 features per frame
TFLite Export
255 KB · on-device · no internet
BSL Label + Confidence
HELLO · 97%
Under The Hood
92.45% accuracy. 255 KB model. Runs on your phone.
EBITH’s BSL recognition pipeline extracts hand and pose landmarks from video frames using MediaPipe, feeds 30-frame sequences into a custom GRU network trained on 9,400 cleaned sequences across 93 signs, and exports a 255 KB TFLite model that runs entirely on-device — no internet connection, no data leaving the user’s device, no latency.
Built for real-world use, not laboratory conditions.
- Test accuracy
- 92.45%Test accuracy
- Model size
- 255 KBModel size
- Training sequences
- 9,400Training sequences
- Network latency
- 0 msNetwork latency
The Gap We Are Closing
Existing solutions cost £2,500–£6,000. EBITH aims to deliver this on any smartphone.
What the market offers today
OrCam MyEye: limitation.£4,000–£6,000
Hardware-dependent — a separate device to buy, charge, and carry.
Envision Glasses: limitation.£2,500+
Single purpose — reads text well, but does not bridge conversation.
Most sign language apps: limitation.ASL-focused
Trained on American Sign Language. BSL is a different language, not a dialect.
What EBITH does instead
Mobile-first: advantage.
Works on any Android or iOS device the user already owns. Nothing extra to buy.
BSL-specific: advantage.
Trained on British Sign Language from the ground up — 93 signs and growing.
On-device: advantage.
Full privacy, offline capability, zero latency. No frame ever leaves the phone.
The cost of assistive technology is itself a barrier. A tool that only reaches people who can find £4,000 has not solved the problem — it has priced it.
Built For Scale
The people who need this most are reached through institutions.
NHS & Healthcare
Deaf patients communicating with hearing staff without interpreters. On-site, real-time, private.
No patient data leaves the device — a requirement, not a feature.
Schools & Special Needs
Children with visual or hearing impairments gain independence tools their school can actually afford.
Deployable on existing school tablets, no per-seat hardware budget.
Government & Transport
Public space navigation for visually impaired citizens. European Accessibility Act compliance ready.
EAA obligations took effect across the EU in June 2025.
Where We Are Going
Roadmap
- Q2 2025Complete
BSL model trained
92.45% accuracy across 93 signs, exported to a 255 KB TFLite model.
- Q3 2025In progress
Flutter app MVP
BSL recognition, object detection, and speech-to-text in one build.
- Q4 2025Planned
UK pilot
Three special needs schools, measuring real classroom independence gains.
- Q1 2026Planned
Innovator Founder Visa · Series A
Formal UK entity, grant funding secured, first commercial partnerships.
From The Lab
What we are learning, written down.
Research
How We Achieved 92.45% BSL Recognition Accuracy on a 255KB Model
A full walkthrough of the EBITH British Sign Language pipeline — landmark extraction, GRU architecture, the dataset contamination bug that gave us a fake 99%, and how we shipped a 255KB on-device model.
8 min
Market Research
The UK Assistive Technology Market in 2025: Where the Gaps Are
Two million deaf people, 340,000 registered blind, a multi-billion-pound market, and a legal deadline that landed in June 2025 — yet British Sign Language remains structurally underserved. A map of the gap EBITH is building into.
6 min
Design & Ethics
Why Building for Both Blind and Deaf Users Forces Better Design
Designing one product for users who cannot see it and users who cannot hear it removes every lazy option. What is left is a stricter, simpler interface — and a set of ethical questions that most AI assistive tools never ask out loud.
5 min
Work With Us
Supporting accessibility shouldn’t require a grant committee. But if you’re one — we’d love to talk.
EBITH is actively seeking UK Innovate UK grants, US NSF accessibility funding, and NHS innovation partnerships. We have the technology. We need the runway.
Or write directly to hello@devsandvisuals.com