Projects
Work delivered through Oravont Systems LLP — national AI challenges, Navy programmes, academic collaboration, and open research artifacts.
Capability overview (PDF)End-to-end deep-learning system for underwater acoustic event detection, multi-class classification and vessel re-identification, operating directly on raw passive-sonar waveforms through a learned, multi-scale selective-kernel front end. Delivered strong, well-calibrated classification across acoustically diverse classes under India’s national-security AI Grand Challenge (NCIIPC / Startup India).
Technical partner building the deep-learning core for automated recognition of vessels from passive-sonar signatures: compact acoustic fingerprints enabling match-or-flag-as-new decisions, designed for in-house, frozen-model deployment with no data egress and no retraining.
Open documentation — disc5-training: model architecture, training methodology and evaluation protocol · disc5-reid: the enrolment and identification application.
Research collaboration in formation with IIT Delhi extending acoustic vessel re-identification through deep metric learning and physics-based acoustic-propagation modelling, in association with a DRDO/NSTL programme.
LLM + retrieval-augmented-generation (RAG) solution for knowledge extraction in naval design and construction, submitted as Principal Investigator under iDEX open challenges.
iDEX proposal for an Active Noise Control silencer for submarine exhaust using CMAC-based control — a direct acoustic-stealth application.