Founding Engineer — AI Infrastructure Optimization

Paralleliq; USA; Remote; USD 140.000 - 190.000; Vollzeit

Benefits dieser Stelle

Equity

Zusammenfassung

Paralleliq is seeking an experienced backend engineer to join an early engineering team building software for diagnosing and improving GPU cluster efficiency for AI inference. You will work on telemetry collection, rules for remediation, and orchestration of durable workflows while supporting on-prem deployments.

The role emphasizes deep knowledge of vLLM internals, Kubernetes Operators, and observability tools, with equity offered for a fast-moving startup environment in the United States.

Qualifikationen

• Five+ years of backend or platform engineering with Python.
• Strong Kubernetes proficiency and production GPU workload experience.
• Deep understanding of vLLM internals: continuous batching, PagedAttention, KV-cache management, sequence-length behavior, V1 engine.
• Experience with inference systems such as SGLang, disaggregated serving, caching, speculative decoding, and GPU sharing/partitioning.
• Observability experience with Prometheus, DCGM telemetry, Kubernetes events, and inference-framework metrics.
• Ability to infer root causes and remediation from cluster signals.
• Experience with durable workflow orchestration and the Kubernetes Operator pattern (CRDs).
• Bachelor's or master's degree in computer science; comfortable across the stack in a fast-moving startup.

Aufgaben

• Extend a scanner to collect telemetry from inference frameworks, GPU tools, Kubernetes, and network operators.
• Build rules linking patterns to likely causes and remediation steps.
• Account for production serving designs, including multi-stage architectures, caching, GPU partitioning, and speculative decoding.
• Develop the control plane and stateful remediation workflows with safety and auditability.
• Collaborate with customers on on-prem deployments, constraints, and validating rules against live telemetry.
• Own work across telemetry collection, backend services, and the operator dashboard.

Kenntnisse

Python programming

FastAPI

SQLAlchemy

Kubernetes

GPU workloads

Telemetry/Observability

Root cause analysis

Kubernetes Operator patterns

Ausbildung

Bachelor's or master's in computer science

Tools

Prometheus

DCGM telemetry

SGLang

CRDs & Custom controllers

vLLM internals concepts

Jobbeschreibung

Role overview

Join an early engineering team building software to diagnose and improve the efficiency of GPU clusters running AI inference. The system gathers model-aware telemetry, distinguishes between different causes of poor performance, and turns findings into safe, auditable remediation. The role spans scanner development, optimization logic, workflow orchestration, customer deployments, and the operator interface.

Responsibilities

• Extend a scanner to collect telemetry from inference frameworks, GPU monitoring tools, Kubernetes, and network operators.
• Build rules that connect patterns such as KV-cache pressure, network saturation, workload misplacement, and thermal throttling to likely causes and remediation steps.
• Account for production serving designs, including multi-stage architectures, caching, GPU partitioning, and speculative decoding.
• Develop the control plane and stateful remediation workflows, making fixes safe and auditable.
• Work with customers on on-premises deployments, constraints, and validating rules against live telemetry.
• Own work across telemetry collection, backend services, and the operator dashboard.

Requirements

• At least five years of backend or platform engineering experience using Python, with FastAPI and SQLAlchemy experience.
• Strong Kubernetes proficiency and experience building or operating tools for production GPU workloads.
• Deep understanding of vLLM internals, including continuous batching, PagedAttention, KV-cache management, sequence-length behavior, and its V1 engine.
• Working knowledge of inference systems such as SGLang, disaggregated serving, caching, speculative decoding, and GPU sharing or partitioning.
• Observability experience with Prometheus, DCGM telemetry, Kubernetes events, and inference-framework metrics.
• Ability to infer root causes and suitable remediation from cluster signals.
• Experience with durable workflow orchestration and the Kubernetes Operator pattern, including CRDs and custom controllers.
• Bachelor's or master's degree in computer science; comfort working across the stack and in a fast-moving startup environment.

Benefits and work setup

• Full-time, remote within the United States. Compensation includes equity.