About us:
TransPerfect, a recognized leader in translation software with a vibrant start-up spirit, is seeking a creative and passionate Backend Developer to join our innovative Artificial Intelligence (AI) team. As part of this division, you will have the opportunity to shape the future of AI in a global organization. From its beginnings over 10 years ago and the creation of its first machine translation models, the AI team has become a core driver of the company's innovation in machine translation, generative AI, natural language processing, and automation.
We are looking for an experienced backend developer who is excited about pushing the boundaries of technology and making a lasting impact within the AI space. You will be part of a diverse, global team of professionals across the USA, Spain, Portugal and India. If you are passionate about robust and scalable solutions that bring AI to users, this is the role for you.
About the Role:
As a Backend Developer, you will help us solve the "last mile" of document processing: converting complex, unstructured PDFs into perfectly formatted, editable .docx files. The goal is not just to extract text, but to recreate the visual and structural intent of the original document—including nested tables, multi-column layouts, font hierarchies, and styling.
You will lead the research and implementation of our document conversion pipeline. This is a hybrid role requiring you to be both a strategic decision-maker (staying on top of the existing tools) and a hands-on developer (combining engineering and AI skills).
You will be in charge of:
· Comparative Analysis: Perform a deep-dive evaluation of commercial (ABBYY, Adobe, AWS Textract) vs. open-source/AI-native (Mistral OCR, Docling, Nougat, LlamaParse) solutions.
· Benchmarking: Establish metrics for "format fidelity" to objectively measure how well a tool recreates headers, footers, tables, and styles.
· Pipeline Development: Build a Python-based workflow that integrates OCR engines with document generation libraries (like python-docx or Pandoc).
· AI Implementation: Explore and fine-tune Vision-Language Models (VLMs) or LayoutLM-style architectures to improve structural recognition.
· Optimization: Solve specific edge cases such as rotated text, low-resolution scans, and complex mathematical notation.
Technical Requirements
· Python Mastery: Expert-level Python skills with experience in OpenCV, PyMuPDF, and python-docx.
· OCR/Document AI: Deep familiarity with Tesseract, PaddleOCR, and modern Transformer-based document models (LayoutLMv3, Donut, or Nougat).
· Format Expertise: A "pixel-perfect" mindset—understanding the nuances of XML-based document formats (OOXML).
· LLM Integration: Experience using GPT or Claude models for layout correction and semantic cleanup.
· Architectural Vision: Ability to decide when to use an "off-the-shelf" API versus when to build a custom PyTorch/TensorFlow pipeline.
Nice to Have
· Experience with Pandoc AST (Abstract Syntax Trees) for format conversion.
· Background in DTP, Typography or Graphic Design.
· Contributions to open-source OCR or PDF manipulation projects.