Bengaluru, Karnataka, India
I like writing good code to solve cool problems.
- Reduced MTTR by 50% by building a JVM Diagnostics Platform enabling on-demand diagnostics across 16 Oracle Cloud environments to speed up incident resolution; developed AI agent skills to integrate the entire workflow into our AIOps workspace. - Eliminated manual config drift risk across 47 Oracle Cloud Targets by architecting an OAuth2-authenticated Helm rollout tool that automated multi-region configuration management for all Observability Server Instances. - Unified fragmented AIOps tooling by refactoring bootstrap, credential management, k8s wrappers, MCP, and diagnostics features to integrate our division’s workspace into Oracle’s cross-division AIOps platform, delivering a seamless workstation experience. - Assessed production-readiness of two Oracle Cloud Services against 37 security and reliability controls, uncovering 33 failures spanning pod security, network policy, observability, and resource configuration. - Partnered directly with end-users to pilot our AIOps tool, gathering feedback that drove 21 targeted bug-fixes and enhancements
- Expanded AI Gateway capabilities by engineering support for multi-provider image manipulation, enabling diverse media processing workflows. - Enforced enterprise-grade security by implementing end-to-end identity management and team mapping, ensuring compliance with strict provisioning and access control standards. - Optimized system performance and resilience by deploying prompt caching, proxy support, and granular guardrail error handling within the AI Gateway. - Stabilized model cost tracking across the Kubernetes Controller and Server through targeted reliability engineering, ensuring accurate resource usage monitoring. - Broadened integration ecosystems by developing the LmChatTruefoundry node for n8n, exposing the AI Gateway as a native LangChain component for low-code automation.
Enabled proactive adjustments to meet customer commitments by engineering a real-time widget for early visibility into manufacturing shortfalls, leveraging Angular, RxJS, and Highcharts, and ensuring application reliability with comprehensive unit and Cypress end-to-end testing.
Worked extensively on creating efficient image generation diffusion models and explored various techniques such as knowledge distillation and rectified flow. Slashed inference from over 1000 steps to just 4 by engineering a lightweight image generation diffusion model (12.3MB) via knowledge distillation and rectified flow, and achieving competitive image quality on a severely limited compute budget.
My HPCC internship focused on harnessing the power of NLP to unlock biomedical knowledge. I built a knowledge graph, like a map of linked genes, diseases, and drugs, extracted directly from scientific papers. Using tools like BERN2 and nltk, I automated the conversion process, making knowledge graph construction for biomedical use more accessible and cost-effective. This structured map connects scattered information, empowering researchers and clinicians with deeper analysis and the potential for personalized treatments.