Machine learning,engineered intoproducts.
Vikolo builds machine-learning systems, the services that run them, and the apps people use every day. Research to production, in one team.
Speaker 100:02
Can we move the design review to Thursday at three?
Speaker 200:06
Thursday works. I'll send the updated deck tonight.
Streaming speech recognition that keeps up with live audio. Illustrative example.
What we build
One team, from model
to interface.
We work across the whole stack: data, models, the services that run them and the apps on top. What we research is what ships.
Machine learning
Speech and audio systems that hold up outside the lab: speaker recognition with neural voice embeddings, Arabic speech recognition and audio quality gating.
- Speaker recognition
- Speech recognition
- Audio processing
Applied AI
Language and vision models grounded in named sources, with guards that keep every answer cited and checkable against the original text.
- Retrieval and semantic search
- Multimodal vision
- Streaming translation
APIs & infrastructure
Containerised services and model endpoints, real-time streaming and the pipelines between them, deployed on managed cloud and measured in production.
- REST and streaming APIs
- Python and TypeScript services
- Cloud deployment
Native apps
Swift and Kotlin apps built for each platform, with Live Activities, widgets and offline reading where they belong.
- Swift and SwiftUI
- Kotlin and Jetpack Compose
- Web
Products
RecitID
Qur'an identifier and live sermon translation
Hear a recitation. Know the surah, the verse and the reciter.
RecitID names what's being recited, often within seconds. It translates the Friday khutbah live, reads Arabic from a photo, and carries a 604-page Madani Mushaf. Native on iOS and Android.
- Reciters recognised by voice
- 250+Reciters recognised by voice
- Verses indexed for recognition
- 6,236Verses indexed for recognition
- Reciters to listen to
- 120+Reciters to listen to
- Translation languages
- 40+Translation languages
Engineering
Measured in production,
not in a demo.
How we build systems like RecitID: tested on audio they have never heard, guarded at every step, and designed to keep working when a dependency doesn't.
- CaptureOn-device quality check
- CleanFiltering, loudness normalisation
- Speech recognitionArabic speech to textVoice embedding250+ voice profiles
- Match6,236 verses, confidence scored
- AnswerSurah, verse and reciter
Parallel pipeline
One recording, two models at once. Speech recognition and voice embeddings run side by side, then meet at verse matching.
Reciter recognition
Top-1 accuracy on held-out clips
Evaluation
Scored on 10-second clips the model has never heard, in clean, noisy and phone-quality conditions.
Internal evaluation, September 2026, against the live service. 168 of 254 reciter profiles measured.
Guardrails
Language models never write verse text themselves. Answers are grounded in named sources and carry citations.
Graceful fallback
If live streaming is unavailable, transcription switches automatically to a chunked pipeline.
Retrieval
Semantic search across every verse of the Qur'an and 76,309 hadith from 17 collections.
Built with
- Swift
- SwiftUI
- Kotlin
- Jetpack Compose
- Python
- FastAPI
- TypeScript
- Next.js
- Node.js
- Docker
Company
We build what we ship,
and we run what we build.
Vikolo Dynamics is a technology company. We design, build and operate the systems behind our own products, and take on engineering work for organisations that need the same depth: machine learning, AI features, APIs and native apps.
- Company
- Vikolo Dynamics Ltd
- Founded
- 2025
- Product
- RecitID
- Serving
- Worldwide
- Contact
- hello@vikolo.com
Have something
hard to build?
We typically respond within 24 hours.