Jobs at Positron AI

Own rapid-response experiments and integration across AI hardware, software, and customer teams. Run LLM inference workloads on simulators and hardware emulators, profile performance, analyze counters and traces, and explain bottlenecks. Support host-side runtime and driver integration, then convert investigations into reusable models, scripts, experiments, and regression tests. The role requires strong systems programming, hardware proximity, Linux proficiency, and the ability to deliver repeatable results from loosely defined requests.

Build analytical and simulation-based performance models for large-scale AI inference deployments on Positron hardware. Model parallelism, networking, batching, latency, throughput, and mixed accelerator configurations; validate projections against hardware measurements; calibrate models as systems evolve; and create customer-facing sizing and performance-estimation tools. Partner with customers, product, and chip-modeling teams to compare deployments, explain performance drivers, and communicate assumptions and confidence levels.

Build the developer platform enabling AI coding agents to operate Positron AI’s chip simulators. Responsibilities include designing MCP servers, APIs, CLIs, agent workflows, CI guardrails, golden-reference testing, static analysis, reproducible runs, provenance tracking, and evaluation systems. The role partners with architecture and engineering teams to automate model bring-up and ensure agent-produced changes are verified before integration.

Design and own multi-chip distributed simulation infrastructure for AI inference accelerators. Build host-to-chip and chip-to-chip simulators, model communication timing and failure modes, compose simulations across fidelity levels, scale runs to full racks, and correlate results with emulation and hardware measurements. Collaborate with chip modeling, SDK, and runtime teams while developing primarily in Rust, C/C++, and Python.

Optimize end-to-end LLM inference performance on custom FPGA and ASIC accelerators. Responsibilities include profiling latency and throughput, tuning hardware-accelerated attention, optimizing distributed inference and KV cache data movement, building benchmarks and regression tests, and debugging across applications, runtimes, drivers, firmware, and hardware. The engineer will also influence accelerator architecture based on measured performance and support production deployments.

Designs and implements Go-based control-plane software for custom AI inference hardware. Responsibilities span system management APIs, Linux services, hardware interfaces, diagnostics, observability, security, firmware updates, distributed systems, and workload management. The role requires hands-on debugging across application, operating system, firmware, and hardware boundaries, plus technical leadership, mentoring, architecture, and production ownership. Regular on-site lab work includes server racking, system bring-up, hardware troubleshooting, and documentation.

Define multi-generation hardware/software architectures for AI inference silicon, including boot, memory, DMA, scheduling, security, telemetry, and host/device interfaces. Integrate firmware, Linux, runtimes, simulation, and developer tooling; lead architecture reviews; resolve cross-team technical issues; prototype high-risk interfaces; and mentor senior engineers. The role requires deep systems expertise, strong C or C++ skills, and architecture-level ownership across pre-silicon, silicon bring-up, and production software.

Owns AI-assisted engineering across accelerator software teams, embedding agentic workflows into coding, debugging, triage, testing, documentation, CI, and silicon-debug processes. Builds shared agent tooling, maintains evaluations, measures productivity gains, and establishes standards for reliable AI-native development. Requires expert experience with Codex, Claude Code, LLM APIs, evaluations, and adopted developer tooling, plus strong systems or infrastructure credibility.

The Senior Software Engineer will develop high-performance software for executing open-source LLMs on custom hardware, focusing on optimizations and efficient libraries primarily in C++.