What I work on
- Agent pipelines → source, fix, test, ship real code
- ML infra and debugger tooling → inference optimization, CUDA scheduling
- Trustworthy AI → federated learning, zero knowledge proofs, explainability
Merged PRs, this month
| Repo | What it was |
|---|---|
| NVIDIA/TensorRT-Model-Connect | fix(t5): check cudaMalloc status for cross-attention and encoder mask buffers |
| NVIDIA/TensorRT-Model-Connect | fix(marian): check cudaMalloc status for cross-attention and encoder mask buffers |
| NVIDIA/TensorRT-Model-Connect | fix(m2m_100): check cudaMalloc status for cross-attention buffers |
| tinyhumansai/openhuman | Fall back to CIM when wmic is unavailable |
| NVIDIA/gpu-driver-container | Guard unzboot copy by arch in ocp_dtk_entrypoint, amd64 doesn't have the binary |
| NVIDIA/TensorRT-Model-Connect | fix(bark): check cudaMalloc status before committing sampler capacity |
| NVIDIA/TensorRT-Model-Connect | fix(bart): check cudaMalloc status for cross-attention buffers |
| NVIDIA/TensorRT-Model-Connect | fix(whisper): check cudaMalloc status for cross-attention buffers |
| microsoft/fluentui | fix(react-utilities): add loading to img element property whitelist |
Background
- Backend and automation infrastructure for a data-driven marketing platform → real recurring revenue
- Cerebras Fellowship → paper to deployable product on Cerebras hardware
- Deep Learning Research Intern, IIT Roorkee → SegFormer and CFAN work, 82% mIoU
- AI Intern, Golden Bird Healthcare → multimodal clinical documentation system, 82% Dice score
- Zerobase Startup School (top 20, Korea) · Google for Startups x Antler