Intro
I'm finishing a Master's in Computer Science at the University of Southern California (graduating December 2026, 4.0 GPA), focused on operating systems, distributed systems, and hardware for machine learning. I did my Bachelor's in Computer Science at PES University, Bangalore.
This past summer I was a Software Engineer Intern at Microsoft, working on Microsoft Defender's attack signature infrastructure. Before grad school I spent two years as a Software Engineer at Epsilon building cloud services on AWS and Azure. At USC I do research in the Network Systems Lab with Prof. Seo Jin Park on lease management frameworks for rapidly scaling clusters up and down.
Experience
Software Engineer Intern · Microsoft
May 2026 – August 2026 · Washington
- Implemented an optimized distributed index for Microsoft Defender attack signature verification to retrieve sample metadata efficiently. Optimized ingestion in Service Fabric and integrated with Cosmos Stream for uniform data access.
- Built an SRE agentic AI solution for Signature Release under Microsoft Threat Protection to automate root-cause analysis of operational issues for on-call engineers, integrated with IcM, ADO and Geneva.
Software Engineer · Epsilon
August 2022 – August 2024 · India
- Led migration of the product to 10+ serverless environments on AWS Lambda with provisioned concurrency, significantly improving cost-effectiveness.
- Maintained 12+ REST APIs in ASP.NET / .NET Core / C# connecting the database, front end and a rules engine.
- Automated cloud infrastructure as code with Terraform and Azure DevOps, improving CI efficiency.
- Built pipelines for storing and retrieving PEM keys for AWS EC2 tenants, cutting troubleshooting turnaround time.
- Introduced API test suites in ReadyAPI and unit tests in MSTest, raising code coverage by 21%.
- Mentored two new hires and interns on the product scrum team.
Researcher · Network Systems Lab, USC
Prof. Seo Jin Park
- Researching lease management frameworks that enable rapid scale-up and scale-down of clusters for systems research and flash-burst computing.
Machine Learning Engineer Intern · Prodian Infotech
June 2020 – August 2020 · India
- Used machine learning (Python, NumPy, Scikit-learn) to identify the factors most influential in wins and losses of domestic cricket matches in India.
- Trained and compared regression, gradient boosting and LIME models to cross-reference results and reduce bias by 5%.
Skills
Languages: C#, Rust, C, C++, Python, SQL, Unix Shell, CUDA
Cloud: AWS, Azure, Terraform
Frameworks & Tools: .NET Core, ASP.NET, gRPC, PyTorch, TensorFlow, NumPy, Scikit-learn, STL, GDB, Git, Postman
Publication
K. M. Meda, S. Subramanian, A. Bennur, S. Vijay and N. S. Kumar, "Teaching Aid For Program Optimization", 2021 International Conference on Smart Generation Computing, Communication and Networking (SMART GENCON).
Projects
GPU-Accelerated Network Intrusion Detection Engine
An Aho-Corasick multi-pattern matching engine in CUDA C++ that scans network packets against attack signatures in parallel at 80+ Gbps on an RTX GPU. A two-phase DFA build (trie insertion plus BFS failure-link completion) gives O(N) scanning regardless of signature count, and the kernel cooperatively loads the DFA into shared memory to cut global memory round-trips.
CUDA · C++ · GPU
LLM Inference Optimization (Llama 3.2-1B)
A high-performance inference codebase for Llama 3.2-1B with INT4 weight quantization and Mixture-of-Experts to shrink the model footprint and latency on A40 GPUs. Implemented LoRA and slice modes for MoE and evaluated trade-offs between perplexity, next-token accuracy and throughput.
PyTorch · CUDA · LLMs
Weenix: A UNIX Kernel From Scratch
Processes, threads, a virtual file system and virtual memory for a kernel based on 6th edition UNIX, written in C. Includes bootstrapping, a polymorphic VFS with reference-counted vnodes and file descriptors, and copy-on-write fork using shadow objects and shadow walking.
C · Operating Systems
Raft Consensus for Distributed State Machine Replication
Leader election and Raft log replication in C++ over gRPC, with a RIFL layer providing exactly-once semantics for strong consistency, safety and liveness.
C++ · gRPC · Distributed Systems
Kite-Shell
A Unix shell in C++17 on POSIX system calls with a system-aware prompt, custom tokenization and full process management. An AI-assist subsystem built on llama.cpp translates natural language into commands and corrects errors by injecting shell state into the prompt.
C++17 · POSIX · llama.cpp
Coptimizer: Automated Tool For C Source Optimization
Automated source-to-source optimization of ANSI C code using Python Lex-Yacc, applying 14 techniques based on Jon Bentley's rules that complement compiler optimizations. Presented as "Teaching Aid For Program Optimization" at IEEE SMART GENCON 2021. Read the publication.
Python · Compilers · Performance
YACS: Yet Another Centralized Scheduler
A centralized job scheduling framework for distributed systems supporting 3+ big-data scheduling algorithms, boosting throughput by 30%+. Workers communicate over a socket protocol for real-time heartbeats and dynamic task allocation.
Python · Distributed Systems
Generic Splay Tree
A generic splay tree container in C++ for caching user-defined data, reaching up to 90% hit rate in web caches, with a bidirectional iterator so it plugs into existing STL algorithms.
C++ · Generic Programming
ParaTest
A multithreaded tester for C++ executables written in Rust. Runs executables against a large number of precompiled test cases efficiently and prints a test summary.
Rust · Concurrency
Check out my other projects on my GitHub profile!
Contact
I'm looking for full-time software engineering roles starting 2027. The best way to reach me is by email at sriramsub7@gmail.com.