Hi, my name is
Anubhab.
I
I'm a researcher bridging the gap between Deep Learning Algorithms and Specialized Hardware.
I focus on Hardware-Software Co-design and Deep Learning Compilers to unlock peak performance on Sparse Accelerators.
I am actively seeking PhD opportunities in Deep Learning Compilers & Hardware-Software Co-design.
About Me
My research focuses on Deep Learning Compilers, High Performance Computing, and Hardware-Software Co-design for neural networks on heterogeneous hardware. I specialize in compiler optimizations for resource-constrained environments like Edge AI and for specialized hardware such as Sparse Accelerators.
I leverage compiler infrastructures like MLIR to automatically generate efficient code. By designing modular compiler passes—from IR optimizations to backend-specific code generation.I aim to bridge the gap between high-level abstractions and silicon.
Where I've Worked
Research Assistant
BrainSeek Lab, IIT Madras
Developing compiler pass pipelines in MLIR/LLVM to map sparse tensor operations to RISC-V vector extensions for Edge AI accelerators.
- Designed the RICE Dialect: A custom MLIR dialect for optimizing Transformer workloads on RISC-V.
- Built an instruction-scheduled ViT inference pipeline tailored for RISC-V architectures.
- Improved LLVM/MLIR pass order by analyzing phase-ordering effects on code quality.
- Yielded 1.5–2x speedups over scalar baselines for Transformer workloads.
Research Assistant
PACE Lab, IIT Madras
Lead Designer of "Morphling": A multi-backend DSL synthesizer for GNNs that decouples algorithm specification from hardware scheduling.
- Achieved speedups of 17x over PyG and 6.5x over DGL (geometric mean) on large-scale datasets.
- Engineered a sparsity-aware execution engine reducing memory traffic on irregular workloads.
- Implemented custom kernels (CUDA/OpenMP) and a distributed partitioner (MPI) with non-blocking training.
Collaborator
CNERG Lab, IIT KGP
Contributed to an NLP project for detecting online hate speech using Transformer-based models.
- Fine-tuned RoBERTa through rigorous hyperparameter optimization.
- Implemented oversampling techniques to address class imbalance.
Data Engineer
Skuad Labs (Remote)
Designed automated ETL pipelines for large-scale structured/unstructured datasets, improving data availability for analytics teams.
Software Intern
Raja Software Labs
Optimized frontend performance for LinkedIn web components using Ember.js.
B.E. in Computer Science
Visvesvaraya Technological University
Projects
RICE Dialect for Torch-MLIR
A custom MLIR dialect for optimizing Transformer workloads on RISC-V. Implements high-level ops like Attention and SwiGLU with lowering passes to Linalg and vector intrinsics.
LLVM Custom Compiler Passes
Implemented core compiler analyses including Range Analysis (abstract interpretation), Andersen-style Pointer Analysis, Dependence Analysis for loop parallelization, and automatic array bounds checking.
MobileNet Compression Pipeline
End-to-end compression for MobileNet V2. Reduced model size by 7.3x (11.4MB to 1.56MB) via magnitude pruning and post-training quantization (W4A8) with <4% accuracy loss.
Static Analysis Visualizer
Implemented the backend for LLVM/Clang Static Analyzer to parse and visually display detailed bug reports and code diagnostics.
Talks & Publications
Publications
Morphling: Fast, Fused, and Flexible GNN Training at Scale
Anubhab, Rupesh Nasre. Submitted to IEEE TPDS, 2025 (Under Review).
A One-Stop DSL for All Your GNN Workloads
Anubhab, Rupesh Nasre. IEEE HiPC Student Research Symposium.
Morphling: A One-Stop DSL for All Your GNN Workloads
Anubhab, Rupesh Nasre. Invited Poster at IISc Computer Systems Workshop.
Talks & Academic Service
Artifact Evaluator
CGO 2026, ACM/IFIP Middleware 2025
Invited Talk: "The Morphling Framework"
IIT Jammu (Upcoming, Dec 2025)
Invited Talk: "Optimized Code Generation"
IIIT Hyderabad (Upcoming, Dec 2025)
Speaker: EduHiPC 2024
Presented on behalf of Prof. Rupesh Nasre
Teaching Assistant
National Supercomputing Mission (NSM) Week, IIT Madras
Awards
Prime Minister's Special Scholarship Scheme; Google Code Jam Qualifier
Skills & Interests
📚 Relevant Coursework
- GPU Programming
- Parallel Scientific Computing
- Systems for Deep Learning
- Program Analysis
- Advanced Data Structures & Algorithms
Research Focus
Compilers
Co-design & Hardware
Languages
DOTA 2
Strategic Addiction
Cooking
Experimental Chef
ML Stack
- PyTorch
- TensorFlow
- PyG
- DGL
- DistDGL
Top 5%
Techgig Code Gladiators
06. What's Next?
Seeking PhD Opportunities
I am actively looking for PhD positions in Deep Learning Compilers, Sparse Accelerators, and Hardware-Software Co-design. If you are recruiting prospective students or would like to discuss my research, please reach out.
Contact Me