Software Engineer (all levels) - High Performance Inference System

Software Engineer (all levels) - High Performance Inference System

Team Missions and Mandates:

• Build and operate real-time, distributed compute frameworks and runtimes to deliver planet-scale inference for LLMs and advanced AI applications at ultra-low latency, optimized for heterogeneous hardware and dynamic global workloads.
• Develop deterministic, low-overhead hardware abstractions for thousands of synchronously coordinated GroqChips across a software-scheduled interconnection network. Prioritize fault tolerance, real-time diagnostics, ultra-low-latency execution, and mission-critical reliability.
• Future-proof Groq's software stack for next-gen silicon, innovative multi-chip topologies, emerging form factors, and heterogeneous co-processors (e.g., FPGAs).
• Foster collaboration across cloud, compiler, infra, data centers, and hardware teams to align engineering efforts, enable seamless integrations, and drive progress toward shared goals.
• Reduce operational overhead, improve SLOs, and make tokens go brrrrrrrrrrr—positioning Groq as the largest inference powerhouse on earth. 🚀

Your code will run at the edge of physics—every clock cycle saved reduces latency for millions of users and extends Groq's lead in the AI compute race.

Signs We Look For:

• You consistently ship high-impact, production-ready code while collaborating effectively with cross-functional teams.
• You possess deep expertise in computer architecture, operating systems, algorithms, hardware-software interfaces, and parallel/distributed computing.
• You've mastered system-level programming (C++, Rust, or similar) with emphasis on low-level optimizations and hardware-aware design.
• You excel at profiling and optimizing systems for latency, throughput, and efficiency, with zero tolerance for wasted cycles or resources.
• You're committed to automated testing and CI/CD pipelines, believing that "untested code is broken code."
• You're deeply curious about system internals—from kernel-level interactions to hardware dependencies—and fearless enough to solve problems across abstraction layers down to the PCB traces.
• You make pragmatic technical judgments, balancing short-term velocity with long-term system health.
• You write empathetic, maintainable code with strong version control and modular design, prioritizing readability and usability for future teammates.
• Nice to have: Experience shipping complex projects in fast-paced environments while maintaining team alignment and stakeholder support.
• Nice to have: Expertise operating large-scale distributed systems for high-traffic internet services.
• Nice to have: Experience deploying and optimizing machine learning (ML) or high-performance computing (HPC) workloads in production.
• Nice to have: Hands-on optimization of performance-critical applications using GPUs, FPGAs, or ASICs (e.g., memory management, kernel optimization).
• Nice to have: Familiarity with ML frameworks (e.g., PyTorch) and compiler tooling (e.g., MLIR) for AI/ML workflow integration.

Early-career engineers: If you've built high-performance systems in school/research and want to dive into cutting-edge hardware/software engineering, we'll invest in your growth to scale with the team.

The Ideal Candidate:

• Initiates (without derailing): Spots opportunities to solve problems or improve processes—while staying aligned with team priorities.
• Builds stuff that actually ships: Values "code in prod" over "perfect slides." Delivers real value instead of polishing whiteboard ideas.
• Is a craftsmanship junkie: Always asks, "How can we make this better?" and loves diving into details.
• Plays to win (together): Believes winning means everyone wins. Aligns goals with teammates and customers because no one succeeds alone.
• Owns it from whiteboard to watts: Takes full responsibility—debug it, deploy it, celebrate with users (or fix it again). Ensures code stays fast, scales well, and takes ownership of outcomes.

This isn't your typical corporate job—it's a mission to redefine AI compute. If you're the kind of engineer who reads ISCA papers for fun and thinks "I can make that faster," this is your playground.

ISCA Papers to Read:

• To learn more about how Groq achieves strong scaling via a software-scheduled interconnection network to maximize inference speed and throughput: A Software-defined Tensor Streaming Multiprocessor for Large-scale Machine Learning
• To learn more about how the exotic, deterministic architecture of each individual chip unlocks massive compute bandwidth on legacy 14nm processes: Think Fast: A Tensor Streaming Processor (TSP) for Accelerating Deep Learning Workloads 

Logistical Requirements:

• Authorized to work in Canada or United States
• Available to work American Eastern or Pacific hours