M.Tech in Computer Science and Engineering
IIT Hyderabad | 2022 - 2025
Spent quality time with machine learning, compilers, and systems programming; occasionally they were kind enough to make sense.
Engineer | Researcher | Life Long Learner
Building the future at the intersection of machine learning, compilers, and high-performance systems.
I am Soumya Banerjee, an engineer by profession. Powered by curiosity, caffeine, and deeply questionable sleep schedule.
Current Focus
- Machine Learning, Compilers and Algorithms
- AI agentic development
- Books, Badminton, Gym and Chess
About Me
Building intelligent systems by day, questionable side projects after midnight.
I currently work as an ML Engineer at Qualcomm R&D in Bengaluru, India, where I focus on ML systems, compiler-oriented optimization, AI accelerators, and high-performance infrastructure. I enjoy solving practical engineering problems and building dependable software that balances performance, scalability, and usability. Alongside low-level systems and AI development, I like crafting polished user experiences, exploring new technologies, and writing about what I learn. Outside coding, you'll probably find me traveling, listening to music, taking photos, or wondering why a bug disappears after restarting everything twice.
C/C++
Machine Learning
Compilers
Python
Linux
Algorithms
LLVM
System Design
Academic Details
Education Timeline
B.Tech in Computer Science and Engineering
West Bengal University of Technology | 2017 - 2021
Built my CS foundation across programming, data structures, algorithms, DBMS, OS, and software engineering; with plenty of debugging sessions for character development.
Higher Secondary (Class XII)
M.P Birla Foundation H.S School | 2015 - 2017
Focused on Physics, Chemistry, Mathematics, and Computer Science; a perfectly normal combo for people who enjoy difficult equations.
Tech Highlights
What I Work On
ML Systems + Accelerators
Working as an ML + Systems Engineer at Qualcomm, enabling PyTorch operators for open-source LLMs on the Qualcomm AI 100 accelerator while building high-performance vectorized kernels using HVX intrinsics and optimizing Multi-NSP execution pipelines for maximum throughput.
Compiler + ML Research
Focused on ML/DL for compilers and program analysis, including binary similarity for vulnerability detection, ML-based compiler inlining, hottest basic block prediction using CFGs, and program classification using branch prediction techniques.
AI Infrastructure + Engineering
Building scalable AI agentic workflows, automation pipelines, and systems tooling to improve engineering productivity, accelerate experimentation, and optimize large-scale development workflows.
Publications
Research Work
VexIR2Vec: An Architecture-Neutral Embedding Framework for Binary Similarity
VexIR2Vec proposes an architecture-neutral approach for binary similarity. It represents binaries using VEX-IR, normalizes the intermediate representation to reduce architecture and compiler variation, learns a vocabulary of IR entities, and uses a Siamese neural network to compare function embeddings.
ML-based Inlining Cost Modeling for Performance
This work explores ML-based cost modeling for performance-driven function inlining. The model learns from historical optimization data and adapts to program behavior, balancing execution speed against resource overhead to make more adaptive, data-driven inlining decisions than traditional compiler heuristics.
Publication Link: TBD
DynVexIR2Vec: A Hybrid Embedding Framework for Binary Similarity
DynVexIR2Vec proposes a dynamic embedding framework that uses execution traces to capture real runtime behavior of binary functions. While static disassembly provides scalable representations, this work investigates whether incorporating dynamic semantics can improve embeddings for downstream tasks such as function searching and algorithmic program classification.
Publication Link: TBD
Images
My Moments
Hobbies & Likes
Beyond Code
Photography
I enjoy capturing street textures, nature, and travel frames whenever I get time.
Writing
I document learnings from books, research, and experiences through short blogs.
Design
I like crafting interfaces that feel minimal, clear, and fast to navigate.
Continuous Learning
Currently investing in books, badminton,finance lessons and chess.
Contact