14 DISC
Design and Development of Sovereign Multi-Modal Foundation Model
Design, develop, integrate, validate and operationalise a fully sovereign multi-modal foundation model stack for the Indian Navy covering model architecture, multi-modal encoder integration (text, image, audio and sensor telemetry), data curation, training and fine-tuning pipelines, RAG over naval knowledge repositories, secure APIs, user interfaces, guardrails, evaluation, red-teaming, observability and lifecycle management. The foundation model shall be the primary deliverable with three deployment variants: (a) quantised models for CPU-only office productivity and RAG-assisted workflows, (b) full-scale GPU- hosted models for high-throughput mission-specific requirements, and (c) task-specific fine-tuned small models for edge deployment including sensor data processing, network management, autonomous operations and swarm coordination. The solution shall support English and major Indian languages, naval terminology and procedures, air- gapped operation, classification-aware deployments and tiered runtime profiles from edge to secure data centre, with reproducible builds, auditable supply-chain control and resilient operation under denied, degraded and disconnected conditions. All source code, training data, training pipelines, CLI tooling and model weights shall remain the exclusive intellectual property of the Indian Navy. Navy personnel shall be trained to independently update, fine-tune and re-deploy models without ongoing vendor dependency. GPU compute for model training shall be sourced through the India AI Mission, which provides free GPU access to Indian firms developing sovereign AI models. The Indian Navy shall facilitate the winner in obtaining authorisation for India AI Mission GPU resources.
