INDRIYA RESEARCH LABS · SPEECH · VISION · AGENTIC AI
We research AI that connects speech, images and video, including multilingual conversations and models for personal computers.
ONGOING MODEL RESEARCH
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 ↗
RESEARCH AND SOFTWARE
Model adaptation, visual understanding and video repair. Explore the research, the prototypes and the questions behind them.
SPEECH
Speech recognition, speaker attribution and spoken dialogue across Indic languages, Arabic and European languages including German, French and Spanish. Our initial programme includes Hindi–English fine-tuning, speaker-attribution prototypes and exploratory dialogue work.
Read about the programme ↗VISION
We investigate image and video understanding alongside controllable generation. Our “Making AI Work” prototypes examine character consistency, object continuity, action timing and selective repair.
READ THE RESEARCH
Dataset preparation, supervised adaptation and comparison with unchanged baselines.
Read the programme note ↗02 / CONVERSATION ANALYSISAnonymous speaker attribution, changes and overlap, with recognition errors assessed separately.
Read the prototype note ↗03 / VIDEO GENERATION AND REPAIRScene constraints, targeted corrections and a proposed comparison with whole-shot regeneration.
Read the development note ↗TURNWISE AI
Turnwise AI is our AI studio for using and evolving models. Start with your recordings, images, video or documents. Eight starter templates apply patterns such as discovery, evidence synthesis and content creation to help you develop, review and refine your work.
Explore Turnwise AI ↗
A SOURCE CHECK
In this fictional job, a meeting says Friday, the proposal says Thursday, and a sketch marks the entrance. Select a source to see what needs checking.
“Please hold delivery until Friday. The price stays the same.”
A later conversation introduces a change.
The dates disagree. Someone needs to confirm the delivery date before the plan is sent.
“Delivery and installation are included in the agreed scope.”
The document supplies a detail missing from the conversation.
The proposal includes installation. Leaving it out would change the agreed scope.
The sketch supplies location context.
The sketch marks the west gate. Check the current access instructions before scheduling the delivery.
This example is prewritten. It runs no AI and does not upload files or take actions.
Read the systems paper ↗CUSTOMER-DRIVEN RESEARCH
Teaching materials, a learning prototype and buyer-discovery exercises help us examine other customer needs. The teaching toolkit is available on Gumroad.
Explore the additional projects ↗FURTHER WORK
Agentic systems: research automation and prototype workflows, alongside the live Turnwise application.
Speech and vision together: our research examines spoken and visual sources, including images, document layouts and video frames. Model adaptation and evaluation remain specific to each task.
Perception: hearing and assistive systems remain a separate research direction. We have no clinical product.
WORK WITH INDRIYA
Tell us about a speech or vision task, or a workflow your team needs. We can discuss a research collaboration or Turnwise access.
Editorial imagery generated with AI.