Member of the Technical Staff - Document Processing & Workflows

The Role We are looking for a Software Engineer with substantial prior experience working with PDFs and PDF-driven applications. PDFs are an odd legacy format: notoriously frustrating to work with, but critically important for understanding people’s finances. Many important businesses that used to run on paper documents now run on PDFs, including bank statements, paystubs, offer letters, and I‑20 proof of F‑1 visa documents. This role is a strong fit for someone who has worked at companies that do OCR, and document understanding driven workflows. You should be pragmatic. You should think less in terms of exploration alone and more in terms of: How will this perform? How will this scale? Is this simple? Is this reliable? You should be an adept user of machine learning, with enough fluency to reason about model errors. You know what ROC, precision, and recall mean. You can reason through over-selection and under-selection, and compare false positives and false negatives against business needs. The primary trait we are looking for is enough technical knowledge to execute without guidance when requirements are clear. You do not need to be a product engineer, but you should be able to prepare PDFs for machine learning steps and intelligently use those outputs to make full‑stack updates to backend workflows that depend on them. You should have a very strong command of Python, and a strong ability to measure service performance and accuracy with systematic metrics using SQL, such as BigQuery. Machine learning and PDF processing often cross the infrastructure boundary in real‑world applications. You should be comfortable debugging Kubernetes pods that are crash‑looping or restarting, and understanding the impact of queueing, memory, disk usage, and CPU usage, without infrastructure being your sole focus. What You’ll Work On ML ops and quality management challenges in PDF processing Building, scaling, and refining Python‑based application code that deals with PDFs and downstream financial data Ensuring PDF processing is as fast as possible, and that machine learning steps are not bottlenecked by server latency, throughput, or non‑ML PDF‑related processing #J-18808-Ljbffr Two Dots