08/27/2026
One year ago, △ ARCED International was mostly an idea.
On 28 August, it turns one. 🎉
In our first year, we launched FolderManifest, InstantHowTo.dev, and SurveyLoopr, and worked with the Commonwealth of Learning.
What excites us most, though, is where the work is heading.
Inside SurveyLoopr, we’re now working on speaker identification for research interviews and FGDs — fine-tuning speaker-diarization models on audio recordings and generating synthetic overlapping speech to better handle real conversations where people interrupt or speak over one another. In our benchmarking exercise, existing models achieved around 75% speaker-attribution accuracy on our data. With our current fine-tuning approach, some experiments have reached 91%. We are not stopping here and working on improving further.
It’s still R&D, but it reflects what we want ARCED International to become: not just a user of technology, but a builder of technology around real problems in research.
And while ARCED International is one year old, the experience behind it isn’t. Much of what made this first year possible grew out of years of research, fieldwork, data, and experimentation through ARCED Foundation.
We put together a small Year One story here:
https://www.arced-international.com/year-one
Year One was about building the foundation. Now, on to Year Two. 🚀
