Upstage

No Training, No Rules—Just Accurate Information Extraction. Tired of spending weeks training information extraction models for every new document type? That’s history. Meet Upstage Information Extract—an intelligent agent that instantly extracts key-values from contracts, invoices, forms, and more with zero setup and no training needed. Interested? Be the first to try it—join the waitlist now! https://lnkd.in/gVNmrBzH

Have a lot of tables? Most document parsing methods struggle with them. We fixed that. Recognizing tables from PDFs shouldn’t mean dealing with ? broken structures, ? missing headers, ? endless manual fixes. That’s why we built Document Parse—an AI-powered table recognition engine that gets it right, the first time. ? Recognizes tables with unmatched accuracy ? Handles multi-section & nested tables effortlessly ? Converts PDFs into structured HTML—ready for LLMs & automation Interested? - Read more: https://lnkd.in/gF4nEshM - See it in action: https://lnkd.in/gs4xgCfz

?? Insurance claims, medical records, unstructured data—healthcare is full of bottlenecks. Upstage is fixing that with AI-powered document intelligence.Trusted by top insurers, Upstage powers 60% of insurance claims in Korea. Now, we’re scaling our AI success to HIMSS Global Health Conference & Exhibition 2025 to transform claims processing and healthcare operations. ?? 2 days left! Let’s connect and dive into the future of AI-powered healthcare. Reach out to Lucy Park, Kasey Kyungsil Roh, or Galina Fendikevich. ?? Visit our exhibitor page: https://lnkd.in/gQ38mtJX #HIMSS2025 #HealthcareInnovation #AIinHealthcare #MeetUsAtHIMSS

Insurance runs on data—but making sense of millions of records? That’s where Upstage changes the game. Upstage powers 60% insurance claims in Korea—now we’re bringing our expertise to ViVE 2025. ?? Join our session | Feb 18, 2025 "What 5 Million Medical Records Taught an $89B Insurer About Cancer Care—And How AI Made It Possible" ??5 million+ medical records processed ??130,000+ claims automated daily ??82% cost reduction We’ll share how insurers are using AI to turn massive data into smarter policies—and what it takes to build AI that delivers real results. ?? Meet us at Meeting Cube #1362 to discuss AI-powered claims processing, prior authorization, and predictive analytics. See you at #ViVE2025! #ViVEvnet #DigitalHealth #HearthcareTransformation #HealthcareInnovation #UpstageAI

Struggling to make sense of charts in your documents? With the new chart recognition feature, #DocumentParse converts complex charts into structured data like HTML, ensuring seamless integration with LLMs. Charts are often overlooked in document parsing due to their complexity, leading to inefficiencies and reduced accuracy. Document Parse solves this by extracting chart data into structured formats, enabling smarter workflows. What sets Document Parse apart: ??Improved accuracy: Achieves 6.49% higher QA accuracy for chart-related tasks than competitors. ??Enhanced efficiency: Processes documents faster, saving time in your workflows. ?? Explore the full feature breakdown in our blog: https://lnkd.in/gkGtCBTm ?? Try it in the Document Parse playground: https://lnkd.in/gEivRjCD

???Announcing #SolarPro — the most intelligent LLM on a single GPU! We’re also thrilled to share that Solar Pro is now available on Amazon Web Services (AWS) Bedrock Marketplace, as revealed during Swami Sivasubramanian’s keynote at #AWSreInvent. Here’s why you need to use Solar Pro: 1?? Solar Pro excels in handling diverse tasks, languages, and domains, with a special focus on understanding structured text such as HTML and Markdown—critical for enterprise applications. Need to process structured text? Solar Pro has you covered. 2?? Solar Pro achieves up to a 64% improvement in Korean and Japanese compared to the Preview version, outperforming numerous local and global competitors and setting a new benchmark for East Asian language capabilities. 3?? From finance to legal and healthcare, Solar Pro is designed to excel in the most demanding domains. ?? Learn more on our blog: https://lnkd.in/g3kJD2Qr ???Try it out at our Playground: https://lnkd.in/gR9Fa_jA ?? Check it out on AWS Bedrock Marketplace: https://lnkd.in/gRWdgRAX

Small Models, Big Results On December 11th, join #SmallCon, hosted by Predibase, to explore the transformative potential of Small Language Models (SLMs). This event brings together top AI leaders to discuss how SLMs are reshaping industries with efficiency, scalability, and cost-effectiveness. Lucy Park(Chief Product Officer) and Kasey Kyungsil Roh(Head of US Business) will present along with an impressive lineup of speakers:? ? Diego Guerra Orozco, GenAI Partnership Lead Meta ? Loubna Ben Allal, SmolLM Lead Hugging Face ? Manjeet Singh, Sr. Director of AI Platforms Salesforce ? Margaret J., Head of Product Mistral AI ? Pavlo Molchanov,?Distinguished Scientist and Manager NVIDIA ? Daniel Hunter, Prev. the Head of AI Harvey ? Abhishek Patnia, Sr. Staff ML Eng Nubank ? Atindriyo Sanyal, CTO and Cofounder Galileo ?? ? Shreya Rajpal, CEO and Cofounder Guardrails AI ? Sudeep Das, Head of AI/ML for New Verticals DoorDash ? Giuseppe Romagnuolo, VP of AI Convirza ? Paul Beswick, Chief Information Officer Marsh McLennan ? Maarten Van Segbroeck, Ph.D., Head of Applied Science Gretel ? Siddharth Ghatti, Enterprise Architect Marsh McLennan ?Arnav Garg, Head of ML Eng Team Predibase ?Devvret Rishi, CEO and Cofounder Predibase Upstage’s Lucy Park(Chief Product Officer) and Kasey Kyungsil Roh(Head of US Business) will also take the stage to share their insights on leveraging small models to solve real-world challenges with scalable, efficient AI solutions. ?? Register now: https://lnkd.in/gHh8zn9H

First Day at AWS #reinvent2024 ! We’re excited to kick off Day 1 at Booth 391 in the Industry Pavilion. ?? Stop by to experience a 30-second glimpse of how Upstage’s AI solutions—like #SolarLLM and #DocumentParse — are shaping the future of work. Meet our leadership team—CEO Sung Kim, CPO Lucy Park, and CTO Hwalsuk Lee—who will be on hand to answer your questions and share how our innovative AI solutions can empower your business. Learn more: https://lnkd.in/gYasrCB2 #reInvent2024 #SolarLLM #DocumentParse

Struggling to align with human preferences efficiently when training your LLM? You’re not alone. That’s why we’re excited to present #sDPO(stepwise Direct Preference Optimization) at #COLING2025 — a more efficient way to train LLMs to align with human preferences with even stronger performance. 1?? The essence: sDPO splits the preference dataset into chunks and performs DPO training step by step, using the aligned model from each step as the reference for the next. This progressive approach creates intermediary models with better alignment, leading to improved final model performance compared to conventional DPO. 2?? The results: Applying sDPO on SOLAR 10.7B + SFT further increases the H4 score to 74.3 showing better performance than using DPO, and it outperformed larger models as well. Want to learn more? Dive into the details here: https://lnkd.in/ghsy6qkM #AI #LLMs #MachineLearning #sDPO #SOLAR

Learn about LLM pretraining pipeline, including depth up-scaling method used by Upstage to train Solar model family, reducing pretraining compute costs by up to 70%.