APPLIED RESEARCH AT INDRIYA

Speech and vision. Research, prototypes and applications.

We research, prototype and build AI systems for speech, vision and agentic workflows. We investigate multilingual speech, image and video understanding, and controllable generation. Agentic workflows support the experiments and connect model capabilities to applications.

The initial six-month speech programme is complete and under internal review. Vision work includes ongoing model investigations and exploratory video prototypes. Agentic systems include research automation, prototypes and the available Turnwise application.

Illustrative scene of three people considering source material and notes together around a laptop
Illustration of a team reviewing source material.

ONGOING MODEL RESEARCH

NVIDIA, Google
and Apple models.

Our ongoing speech and vision work includes NVIDIA Nemotron, Google Gemma and Apple AFM 3, with experiments on a cloud-hosted Mac.

Explore the models and access options ↗

MODELS AND AGENTIC SYSTEMS

Model adaptation
and workflow design.

Speech work spans Indic languages, Arabic and European languages including German, French and Spanish. Vision work examines images, documents and video frames. The initial speech studies and “Making AI Work” video prototypes give us specific tasks for adaptation, error analysis and repair.

Read the research notes and working papers for completed programme work, prototype methods and further evaluation plans.

RELATED RESEARCH QUESTIONS

Related areas
of research.

These related investigations examine source checking, multi-step workflows, learning and assistive systems alongside our core speech and vision work.

RESEARCH AND YOUR WORK

Where the research
may apply.

Start with your situation. The relevant research may sit across more than one area.

FOR BUSINESS OWNERS

Client handovers
and recurring tasks.

Perhaps a client handover means chasing three people, or the same report has to be checked from scratch every week. Knowledge and workflow research can help frame what to capture, what to automate and where your judgment still matters.

A useful starting point is one recurring task, the material it needs and the time your team spends correcting it. Explore an example first, then discuss whether a focused investigation fits.

HOW THE WORK CONNECTS

What we can compare
across these tasks.

A speech system must preserve a spoken term; a vision system must preserve the relevant detail in an image or sequence. Both need ways to expose errors, accept corrections and retain their sources. Teaching, learning and buyer-discovery projects examine these questions in additional settings.

We look for methods that can carry across settings—source checking, explicit uncertainty, useful correction—while evaluating each application on its own terms. A result in one setting is not proof that another will work.

Explore our approach ↗

WORK WITH INDRIYA

Discuss a
research question.

Tell us about your speech or vision task, the available material and the problem you want to investigate. We can discuss a focused prototype or test, with data access and expected results agreed beforehand.

Discuss your research question