Software Engineer, Inference

Pulse; San Francisco (CA); On-site; USD 120,000 - 160,000; Full time

Benefits offered by this job

Competitive base salary

Performance-based bonus

Relocation assistance

Daily meal stipend

Medical, vision, and dental coverage

Job summary

A data infrastructure company in San Francisco seeks a specialist in low-latency inference for OCR and multimodal models. Candidates should have 3+ years in performance engineering or ML systems, with strong skills in Python and exposure to C++ or CUDA. The role involves building inference services, optimizing models, and implementing autoscaling. Competitive compensation including salary, equity, bonuses, and benefits is offered.

Qualifications

• 3+ years in performance engineering or ML systems.
• Strong Python programming skills with C++ or CUDA exposure.
• Experience with GPU profiling and model serving.

Responsibilities

• Build inference services with smart batching and caching.
• Optimize kernels, tokenization, and model graphs.
• Evaluate vLLM, TensorRT LLM, and Triton tradeoffs.
• Implement autoscaling and admission control with clear SLOs.
• Own performance dashboards and capacity planning.

Skills

Performance engineering

Machine Learning systems

Python

C++

CUDA

GPU profiling

Job description

Overview

Pulse is tackling one of the most persistent challenges in data infrastructure: extracting accurate, structured information from complex documents at scale. We have a breakthrough approach to document understanding that combines intelligent schema mapping with fine-tuned extraction models where legacy OCR and other parsing tools consistently fail.

We are a small, fast-growing team of engineers in San Francisco powering Fortune 100 enterprises, YC startups, public investment firms, and growth-stage companies. We are backed by tier 1 investors and growing quickly.

What makes our tech special is our multi-stage architecture:

• Layout understanding with specialized component detection models
• Low-latency OCR models for targeted extraction
• Advanced reading-order algorithms for complex structures
• Proprietary table structure recognition and parsing
• Fine-tuned vision-language models for charts, tables, and figures

If you are passionate about the intersection of computer vision, NLP, and data infrastructure, your work at Pulse will directly impact customers and shape the future of document intelligence.

What we are looking for

• 5 days in-office at our San Francisco office
• Eager to learn and adapt quickly
• Prior startup or founding experience is a plus

About the Role
Specialize in low-latency, high-throughput inference for OCR and multimodal models. Own profiling, batching, and autoscaling across single-tenant and multi-tenant environments.

Responsibilities

• Build inference services with smart batching and caching
• Optimize kernels, tokenization, and model graphs
• Evaluate vLLM, TensorRT LLM, and Triton tradeoffs
• Implement autoscaling and admission control with clear SLOs
• Own performance dashboards and capacity planning

Requirements

• 3+ years in performance engineering or ML systems
• Strong Python, plus C++ or CUDA exposure
• Experience with GPU profiling and model serving

Nice to have

• Experience reducing p95 and cost in production ML systems

Sponsorship
Sponsorship available.

Compensation and benefits
Competitive base salary plus equity, performance-based bonus, relocation assistance for Bay Area moves, daily meal stipend, medical, vision, and dental coverage.