|
I'm a final-year Computer Engineering undergraduate at FCRIT, Vashi ('27) working at the intersection of AI Research, ML Systems Engineering, and High-Performance Computing. I care about what happens beyond the notebookβtaking models and algorithms from theory into low-latency, compiler-optimized, production-grade systems.
|
role: Software Engineer Intern @ InfinityPool
prev: ML Research Intern @ IIT Bombay (WaVE Lab)
club: Former Technical Head @ AIDL FCRIT
grad: B.Tech Computer Engineering '27
highlights:
"π Finalist β AWS Impact X Challenge (IITB)"
"π Published β IEEE ICCES 2024 (SuperPOS)"
"π¦ Building β Halation (Lightroom in Rust)"
focus_2026:
Compiler-aware ML & GPU acceleration
Distributed AI inference & Ray / K8s
Low-latency quantitative & systems engineering |
|
Jun 2025 β Present Β· Navi Mumbai (Hybrid)
|
Jun 2025 β Jul 2025 Β· Under Prof. Nitesh P. Yelve
|
|
Jun 2025 β Jun 2026 Β· Navi Mumbai
|
SuperPOS: Interactive Web-Based Quantum Simulator
|
π¨ HalationLightroom-class photo editor & color grader built from scratch in Rust
|
π Finalist Β· AWS Impact X Challenge (IIT Bombay)
|
|
Agentic software engineering workflow using LangGraph & DeepAgents
|
High-performance systems exploration for model serving
|
β‘ Expand Technical Deep-Dive: ML Compiler & Hardware Optimization Benchmarks
Benchmarked identical Transformer decoder blocks & ResNet50 across PyTorch, TensorFlow, and JAX on NVIDIA GPUs and Google TPUs to evaluate compiler-level gains over eager execution:
torch.compile + Tritonβ 2β3Γ inference/training speedup over eager PyTorchPyTorch + TensorRT (FP16)β Major deployment throughput & GPU-count reductionJAX / Flax + jit (XLA)vs.PyTorch/XLAon TPUs &tf.functiongraph capture
Languages & Systems
AI / ML, Graphics & Data
Also working with: Hugging Face Β· LangGraph Β· Amazon Bedrock Β· ONNX Runtime Β· NVIDIA Triton Β· TensorRT Β· JAX/XLA Β· WGPU Β· TimescaleDB
Backend, Full Stack & Cloud/Infra
Always open to collaborating on AI infrastructure, developer tools, and high-performance systems β let's talk.

