Sohan Subudhi
Turing Scholar pursuing integrated B.S. and M.S. degrees in Computer Science at UT Austin. I work across compilers, systems, and AI infrastructure.
Austin, TX
currently
- playing
- rocket league · pickup basketball
- building
- llm-guided compiler optimization workflows
- learning
- llvm internals · arm/aarch64 · sve
Skills
- languages
- Python · C/C++ · Java · R · SQL · Bash
- systems
- LLVM/Clang · ARM/AArch64 · SVE · Linux · eBPF · Docker · GDB
- cloud/data
- Google Cloud · AWS · PostgreSQL · CI/CD · Redis · GitHub
- ai/ml
- PyTorch · GNNs · LLM agents · RAG · Hugging Face · NumPy · Pandas
Experience
- May 2026 – Aug 2026
Compiler Software Engineer Intern · Qualcomm
San Diego, CA
- Automated compiler flag tuning and C++ rewrites using source-level compiler hints and SVE intrinsics for an ARM CPU target.
- Built an LLM agent that generated optimization candidates, compiled and benchmarked them on device, analyzed code generation, and iteratively refined its approach through measured feedback.
- Developed a RAG pipeline using LLVM documentation, IR embeddings, and code examples to ground optimization candidates.
- Achieved 5–10% performance gains over previous best benchmark compiler flag sets after 5–6 optimization iterations.
- Jan 2025 – May 2026
Undergraduate Researcher · NSF Learning-Directed OS Lab
UT Austin
- Built a closed-loop LLM workflow to automatically design, generate, verify, and evaluate Linux kernel congestion control algorithms for low-delay, high-throughput networking.
- Trained a Graph Neural Network on workload execution traces to model behavior and optimize hardware resource allocation.
- May 2025 – Aug 2025
Software Engineer Intern · UKG
Weston, FL
- Architected the company's first multi-agent AI system with Google ADK, correlating technology scanned by CheckmarxOne and Wiz and automating remediation through Jira and GitHub — cutting vulnerability risk by 23%.
- Fulfilled queries 12× faster, with a two-second average delay, through multi-level Redis caching and custom runtime handling.
- Ensured reliable real-time technology inventory data through ELT pipelines and a GitHub Actions CI/CD workflow.
- Jun 2023 – Aug 2023
Data Science Intern · VRIZE
McKinney, TX
- Automated training of custom time-series models on a self-service platform, integrating PostgreSQL and AWS S3 for data management.
- Reduced model training time by 90% through statistical grid search, stationarity testing, autocorrelation analysis, and AWS SageMaker parallelization.
- Built interactive real-time Plotly dashboards for historical data, forecasts, and predictions.
Projects
-
Object-Oriented Programming Compiler
ARM assembly compiler with OOP, arrays, imports, a custom buddy/fifth-fit heap manager, and optimizations bringing runtimes within 15% of GCC.
c++ · bash -
6-Stage Pipelined Processor
AArch64 subset (misaligned/self-modifying support) with forwarding, dynamic stalls, and speculative branch prediction — 2.1× speedup, CPI 2.8 → 1.3.
verilog · gtkwave -
Waste Detection Model
YOLOv5 satellite-imagery model at 92% mAP for land waste detection. Deployed as a public REST API on Streamlit Cloud.
python · cvat
Education
The University of Texas at Austin
M.S. in Computer Science · Integrated Five-Year Master's
B.S. in Computer Science · Turing Scholar Honors Program · GPA 3.94
Coursework: Honors Computer Architecture, Honors Operating Systems, Honors Algorithms, NLP, Honors Machine Learning I & II, Linear Algebra, Numerical Methods, Co-Design for Datacenters/Cloud, and Distributed Systems.
Awards
- 2025
Top 15, eBay University ML Competition
Five-month national NLP challenge
- 2023
1st Place, Peddie Hacks Hackathon
College division · 278 participants