Applied AI Engineer

Ersilia; San Francisco (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Ersilia is seeking an Applied AI Engineer to turn research into production, focusing on browser agent reliability, document understanding, and inference optimization. You will build systems that improve accuracy and speed, spanning UI interaction, unstructured data parsing, and tool use.

Work with state-of-the-art techniques, design fine-tuning pipelines for customer workflows, and optimize latency with model selection, quantization, caching, and routing.

Qualifications

• Strong Python and ML frameworks, particularly PyTorch.
• Eval-and-metric mindset for production-focused metrics.
• Comfort with messy data and making it useful.
• Track record of shipping end-to-end systems.
• Clear, buzzword-free communication about your work.
• Based in San Francisco or willing to relocate; in-person 5 days a week.

Responsibilities

• Develop production-grade AI systems for UI interaction, document understanding, and tool use.
• Improve latency across the stack via model selection, quantization, caching, and routing strategies.
• Design fine-tuning pipelines for customer-specific workflows.
• Build self-healing systems that adapt when environments change.

Skills

Python

PyTorch

ML frameworks

Evaluation metrics

Data wrangling

End-to-end shipping

Clear communication

San Francisco base

Tools

PyTorch

TensorFlow

Job description

About the Company

We deploy AI that operates computers the way humans do: navigating browsers, processing documents, working through legacy systems. The company works with large enterprises to automate their messiest finance operations, attacking the $300B+ BPO industry built on labor arbitrage that software couldn't touch because people were the product.

About the Role

As an Applied AI Engineer, you'll work on everything from browser agent reliability to document understanding to inference optimization, building systems that make the work more accurate and faster every week. This is a role for someone who thrives on turning research into production, working on problems like pushing to state-of-the-art across core automation capabilities (UI interaction, unstructured data parsing, tool use), building adaptive systems that self-heal when environments change, designing fine-tuning pipelines that learn from customer-specific workflows, and optimizing latency across the stack through model selection, quantization, caching, and routing strategies

Requirements Must-Have

1. Strong Python and ML frameworks, particularly PyTorch.


2. Eval-and-metric mindset. You think in terms of metrics that matter in production, not just benchmarks.


3. Comfort with messy data and figuring out how to make it useful.


4. Track record of shipping. You can describe specific systems you've built end-to-end.


5. Crisp communication about your own work. You can describe what you built in a few clear sentences without buzzwords.


6. Based in San Francisco or willing to relocate, in-person 5 days a week.

Nice-to-Have

1. Experience with RL, retrieval systems, or agent-based systems


2. Cross-stack range: inference optimization, data pipelines, fine-tuning, and model monitoring


3. Published ML papers or significant OSS contributions


4. Lab or research exposure (SAIL, BAIR, MIT CSAIL, similar)


5. Recent applied work on LLMs, browser agents, RAG, or production AI workflows