I'm Veluvolu Deepak

Worked on Tile Size and Loop Order Selection using Machine Learning for Multi-/Many-Core Architectures, where I developed a Data Dependence Graph (DDG) system that automatically generates randomized, yet representative nested loop structures as comprehensive training data for machine learning models.

• This synthetic loop generator enables ML classifiers to accurately predict optimal tile sizes and loop orders by creating thousands of diverse 2-dimensional loop nests with controlled dependency characteristics, array access patterns, and parallelism properties to improve GPU performance.

• Contributed to the “श्रुvaan” research work on a cryptographically grounded model-context protocol for secure, natural, multilingual agent-to-agent communication in mission-critical domains.
• Implemented and experimented with core ideas such as role-aware access control (HKP-style) and verifiability (Proof-of-Protocol) to improve trust and integrity of agent messages.
• Evaluated threat surfaces like prompt injection, schema mimicry, cross-context spoofing, and multilingual ambiguity, and supported prototype iterations based on findings.