A tiered KV-cache system for GPT-2 inference that moves data across local NUMA memory, disaggregated memory, and disk using LRU-based placement under constrained memory.
Python · PyTorch · Cython · NUMA · LLM inference
Building
Infrastructure
at Scale
Hemachandran, username @hemacdn
About
I'm Hemachandran Balaji — a Software Engineer at Amazon, where I build real-time telemetry for Amazon Music. My work sits at the intersection of distributed systems and data: making millions of listening sessions observable, measurable, and reliable.
Across embedded sensors, applied machine learning, and real-time telemetry, I keep coming back to the same questions: where does a system slow down, how does it fail, and how can it be made more dependable as it grows? That is what draws me to systems performance, distributed systems, and infrastructure.
Experience
Software Engineer
Amazon · San Francisco
Building real-time telemetry for Amazon Music's distributed systems.
2025 — Present
Robotics Engineering Intern
Labelbox · San Francisco
Built a low-latency Meta Quest teleoperation system and ROS control and data pipelines for real-time robot manipulation.
Jul 2025 — Sep 2025
GenAI Technical Advisor Intern
Scale AI · San Francisco
Designed and validated Python-based evaluation tasks for generative AI models, helping identify reasoning failures and improve training-data quality.
Jun 2025
Embedded Software Researcher
Lab for AI, Robotics & Automation · UC Davis
Contributed to the design and development of an Arduino-based navigation device for the visually impaired and helped optimize the sensor-update loop, reducing latency by 50%.
Oct 2023 — Jun 2024
Software Engineer Intern
Software Engineering Research Center · IIIT Hyderabad
Built a real-time data pipeline for a pressure-sensitive mat; improved throughput by 50% and cut processing time by 25% with a sparse matrix JSON serialization framework.
May 2022 — Jun 2022
Education
University of California, Davis
M.S. Computer Science
2023 — 2025GPA 4.00
Vellore Institute of Technology
B.Tech Computer Science
2019 — 2023GPA 3.88
Projects
Image super-resolution with a GAN compressed via knowledge distillation — EfficientNet meets PatchGAN, lifting SSIM 12% while cutting inference time 35%.
Python · TensorFlow · GANs
Where should California put its next wind farm? Scraped geospatial wind and cost data, then built an interactive map to surface the optimal sites.
Python · Pandas · Folium · Selenium
AirTouch
A gesture-controlled keypad built on Arduino — ultrasonic sensors, no touch required. Improved gesture accuracy 20% over the baseline approach.
Arduino · C++ · IoT
Skills
Languages
- Python
- Java
- C/C++
- TypeScript
- SQL
- R
Tools & Platforms
- AWS
- Docker
- Linux
- Node.js
- React / Next.js
- Git
Certification
- AWS Certified Cloud Practitioner · Valid through 2027