• Uber's GenAI Invoice Processing System: Reduced manual effort by 2x, cut handling time by 70%, and saved 25–30% cost. Leveraged GPT-4 and TextSense for 90% data accuracy and globally scalable, efficient, highly automated financial ops.
• Shift from Traditional Tools: Replaced RPA and RBS with GenAI to address complexity and inefficiencies. Traditional tools lacked adaptability for invoice formats at Uber's scale, while GenAI offered flexibility to adapt to new formats without manual rule-setting.
• TextSense Platform: A modular, scalable document processing platform as the backbone. Abstracts processes, integrates OCR, LLM extraction, and post-processing with reusable components for rapid onboarding of new formats.
• Global Scalability: Allows flexible scaling across 25+ languages, various formats (including handwritten and scanned), maintaining accuracy and efficiency while handling diverse supplier invoice templates.
• Human-in-the-Loop Review: Combines GenAI with HITL review for high accuracy and human oversight. A purpose-built UI enables side-by-side comparison of extracted data and original PDF, accelerating validation with intuitive design and alerts.
• Model Evaluation: Compared fine-tuned open-source models like Flan T5 and LLaMA 2 with proprietary solutions. GPT-4 delivered superior accuracy across header and line-level fields with minimal tuning. Newer multimodal models like GPT-4o, Claude 3.7, and Llama 4 are available now, but Uber's model evaluation timeline is unknown.