Capt (Dr) Sunil Tyagi, Indian Navy (Retd)

Over 32 years of commissioned service in the Indian Navy (1992–2024), Sunil Tyagi's specialist core was submarine acoustic stealth and underwater acoustics. For six years (2015–2021) he headed the Acoustic Stealth Group of the Indian Navy's Submarine Design Directorate, the group responsible for low underwater noise in submarine design across India's nuclear-submarine programme. In that Navy role, before Oravont, he developed India's first indigenous submarine acoustic model, brought underwater-radiated-noise estimation in-house and institutionalised standardised shock-mounting design methods. Earlier appointments ran from engineer officer at sea to submarine diesel engines at Naval Dockyard, Visakhapatnam; later ones to the Naval Trials and Acceptance Authority and the Centre for Air Power Studies.
He holds a PhD in Mechanical Engineering (Applied Machine Learning) from the Defence Institute of Advanced Technology, Pune (2018). That research produced eight peer-reviewed journal papers and an Indian patent in machine-learning-based fault detection on which he is named inventor (Indian Patent 373427; patentee DIAT Pune), and he taught at DIAT and MILIT, Pune. In September 2026 he published the SKANN preprint on arXiv and filed an Indian provisional patent application on open-set vessel re-identification.
Profile, CV, experience and full publication record: suniltyagi.in · LinkedIn · ORCID 0000-0001-5897-7955 · ResearchGate · GitHub · Hugging Face · WILDLABS · YouTube · X · Google Scholar · Scopus · Semantic Scholar · Wikidata
Working principles
- Physics first. Classical, interpretable features run alongside learned representations. When a model separates two recordings, the physics should show why.
- Honest measurement. Evaluation protocols are designed to resist flattering scores, and limitations are published with the results.
- Sovereign deployment. Frozen models, galleries that grow in-house, no data egress and no retraining — on the user's own infrastructure.
- Licence-clean. Development on open-source corpora and open-source dependencies; every released artefact carries an explicit licence.
- Unclassified by design. Public materials describe methods at headline level. Programme specifics, customer data and evaluation results from sponsored work are not disclosed.
- Open where it can be. Preprints, CC BY artefacts and MIT-licensed tools, so that results can be checked by anyone who cares to.
Engagements
Oravont has worked as technical partner to an industry participant in India's national-security AI Grand Challenge (NCIIPC / Startup India, Problem Statement 12, concluded May 2026) and in iDEX, Ministry of Defence (DISC5 Challenge 10, Indian Navy, Underwater Domain Awareness; the programme is active and Oravont's part is winding down); it has an exploratory research collaboration with IIT Delhi, in association with a DRDO/NSTL programme; and it has submitted iDEX open-challenge proposals as Principal Investigator. Engagements take the form of collaborative R&D, funded research, consultancy and advisory work, and technical training and knowledge transfer in acoustics, vibration and applied machine learning.
- Oravont Systems LLP · registered office C-43, Sector 39, Noida 201303, Uttar Pradesh, India
- Email info@oravontsystems.com · GitHub github.com/Oravont