INDRIYA RESEARCH LABS — PUBLIC COMPANY BRIEF Updated October 15, 2026 IDENTITY Indriya Research Labs, Inc. Founded: 2026 Location: Newark, Delaware, United States Company: AI software and model research What we do: We research, prototype and build AI systems for speech, vision and agentic workflows. Focus: multilingual speech, image and video understanding, controllable generation and agentic systems WORK Turnwise AI — Commercial AI studio. Bring recordings, video, images, documents or a concept. Turnwise helps clarify the goal and proposes models, instructions and tools for the task. Eight starter templates describe reusable ways to work: discovery and definition, evidence synthesis, decisions and actions, coordinated planning, content creation, process capture, learning design and solution design. Each template stands on its own and includes a concrete example. Requirements to Proposal illustrates discovery; Proposal Development illustrates solution design. Develop and review your work, then refine the capability for reuse. Owned and operated by Indriya Research Labs, Inc. Independent license holders operate their product tenants and serve their customers. Third Ray is a license holder and service partner. The current Hindi-language product example is tailored to its intended audience. Marketing: https://turnwiseai.com/ Application: https://turnwiseai.app/ Compare Turnwise: https://indriyai.com/products/turnwise/compare/ MULTILINGUAL SPEECH CAPABILITIES Indic languages, Arabic and European languages including German, French and Spanish. SPEECH RESEARCH PROGRAMME The initial six-month programme of applied speech research and prototype development is complete. Comparative findings are under internal review. Selected prototypes are being hardened for pilot deployment; broader validation and customer-specific adaptation remain ongoing. Qwen3-ASR study: Hindi-English transcription adaptation workflow, dataset preparation, transcript normalization, supervised experiments and adapted-checkpoint comparisons with unchanged foundation baselines. Error analysis examined language switches, specialist terms, numbers and accents, plus general-language recognition after adaptation. pyannote-based conversation analysis: Anonymous speaker attribution, changes and overlapping speech, with configuration experiments and targeted segmentation adaptation. Speaker-assignment and transcription errors assessed separately. Consistency across unfamiliar speakers and recording environments remains ongoing. Moshi: Exploratory spoken-dialogue feasibility study, controlled interaction environment and investigation of local inference and further adaptation requirements. Broader validation and pilot readiness are later milestones. Deployment preparation: AWS GPU training and controlled evaluation; Apple Silicon local inference experiments; checkpoint conversion, compatibility, memory profiling and accuracy after compression. Research: https://indriyai.com/research/models/ VISION RESEARCH AND ONGOING MULTIMODAL MODEL WORK Vision research examines images, document layouts and video frames. Controllable video generation and selective repair form a related programme. NVIDIA Nemotron and specialist speech models, Google Gemma, and Apple AFM 3 are included in ongoing research, with experiments on a cloud-hosted Mac alongside the wider AWS environment. We evaluate open-weight deployment and Apple framework integration, recording runtime support, quality, latency, memory and cloud calls. Models and primary sources: https://indriyai.com/research/models/#platforms Evaluation: https://indriyai.com/papers/evaluation/#platforms VIDEO RESEARCH PROGRAMME Making AI Work sequences support exploratory prototype development in controllable multimodal generation and agent-assisted video production. Work includes character and object consistency, action timing, selective repair, failure documentation and verification procedures. Humans govern creative acceptance. Comparative evaluation and generalization across scenes remain research objectives. Research: https://indriyai.com/research/video/ AGENTIC SYSTEMS Research engineering uses AI-assisted data preparation, experiments, evaluation scripts, error categorization and documentation. Video workflows support references, versions, editing, assembly and quality checks. Humans govern research design, dataset acceptance, interpretation and creative acceptance. Agentic systems also include prototypes and live Turnwise workflows. Research: https://indriyai.com/research/agentic-systems/ ADDITIONAL CUSTOMER-DRIVEN RESEARCH Teacher Copilot — Available. US$19 downloadable AI workflow toolkit on Gumroad for grades 6–12 educators; twelve workflows; uses the educator's school-approved AI tool. Not a hosted AI app or subscription. Learning Studio — In development. Source-grounded learning prototype. Public access is not yet available; the public preview is a recorded example, not live AI. BuyerProof — Research exploration. Buyer-discovery workflows for startups and AI-native founders. A free note-organizing exercise is available; it does not use AI or validate demand. Perception and assistive systems — Research direction. Hearing, surroundings and adaptive feedback. No clinical product, clinical efficacy or regulatory approval is claimed. RESEARCH NOTES AND WORKING PAPERS Adapting speech recognition for Hindi-English conversations Which speaker said which words? Interruptions, corrections and spoken replies One character across a sequence Designing agentic workflows Testing models and workflows https://indriyai.com/papers/ The papers describe speech experiments, prototypes and proposed model and workflow comparisons. Comparative speech findings are under review. FOUNDER Abhijit Kumar Choudhury Professional profile: https://www.linkedin.com/in/abhijitchoudhury Company profile: https://www.linkedin.com/company/indriya-research-labs/ CUSTOMER ORGANIZATIONS Third Ray and Azureiken Technologies are customers of Indriya Research Labs. CONTACT https://indriyai.com/contact/ contact@indriyai.com