Software Development Engineer Intern, Annapurna Labs

At Amazon, our tech isn't just a tool, it's your playground. Our engineers work on scalable systems, cloud services, and customer-facing products that operate at global scale. This is an environment where you learn by building. Whether you're creating cloud-native solutions, optimizing machine learning models, or building products that drive experiences for millions of customers around the globe, your work reaches the world, fast.

You'll take ownership early, collaborate across teams and disciplines, and grow your skills as you tackle new problems. There's no single path forward here, with a chance to explore different directions and shape your journey as you go. This is where ambition meets opportunity and, where the impact of what you build helps define what comes next, for customers and for you.

What's in it for you? An internship at Amazon means real responsibility from day one. You'll build, test, and learn alongside people who want to see you succeed, while making an impact that reaches far beyond campus.

Annapurna Labs designs the custom silicon at Amazon - including Graviton (our server processors), Trainium and Inferentia (our machine learning training and inference accelerators), and the Nitro system that powers modern EC2. We own the full stack, from silicon through the software that makes it work (compilers, runtimes, drivers, and the frameworks that let customers run workloads on our hardware).

Annapurna has software teams spanning the full stack, and we work to place you on a technical team aligned with your interests and strengths. Example teams and focus areas include:

• Compiler development for ML accelerators (Neuron SDK, MLIR, LLVM)
• ML runtimes and frameworks (integration with PyTorch, JAX, Tensor Flow)
• Low-level systems software, firmware, and device drivers
• Performance optimization and kernel development for custom silicon
• Distributed training and inference infrastructure

Once you join, we partner with you and our hiring managers to match you to a team where your background and interests fit best, so your project is something you're genuinely excited to work on.

You'll contribute to real production software, which may include the following responsibilities:

• Designing, implementing, and testing software across the stack - from low-level systems code to compilers, runtimes, and tooling
• Writing efficient, maintainable code in languages such as C, C++, or Python
• Optimizing software performance on custom hardware (Graviton, Trainium, Inferentia)
• Debugging complex issues across the hardware-software boundary
• Collaborating with hardware, compiler, and systems teams to deliver features end-to-end
• Documenting your work and presenting results to the team

You'll be paired with a mentor and manager who will scope your project, support your growth, and set you up to deliver meaningful, measurable impact by the end of your internship.

• Currently enrolled in a Bachelor's or Master's degree program in Computer Science, Computer Engineering, or a related field, with a final graduation date of December 2027 or later
• Programming experience in at least one modern language such as C, C++, Java, or Python
• Coursework or project experience in data structures, algorithms, and software design
• Understanding of core computer science fundamentals: operating systems, computer architecture, or systems programming

• Experience working in a Linux/Unix environment
• Experience working with fast-moving, high-performance teams and driving innovative solutions tailored to unique business environments
• Can show strong verbal and written communication skills
• Pursuing a degree with a focus on systems, compilers, computer architecture, or high-performance computing
• Experience with low-level or systems programming in C/C++ (memory management, concurrency, performance optimization)
• Familiarity with compilers (MLIR, LLVM, TVM), ML runtimes, operating systems internals, or distributed systems
• Exposure to machine learning frameworks and an understanding of how models run on hardware
• Prior…