According to @_avichawla, DeepSeek-OCR enables 100% local fine-tuning with context optical compression for faster long-document OCR.

According to Kye Gomez (@KyeGomezB), MoonViT is a native-resolution Vision Transformer that encodes images of arbitrary size without resizing or padding while preserving efficient batching and large language model compatibility. As reported by the original tweet thread, this architecture targets multimodal pipelines where fixed-size crops degrade detail, enabling enterprise use cases like document understanding, medical imaging, and geospatial analysis that need pixel-accurate features. According to the tweet, maintaining batching efficiency suggests MoonViT can scale inference throughput for production multimodal systems, reducing preprocessing overhead and improving latency. As stated by Kye Gomez, LLM compatibility indicates straightforward integration into vision-language models, opening opportunities for higher-fidelity visual grounding and improved OCR-free parsing in RAG workflows.

According to DeepLearning.AI on X (Twitter), the organization launched a short course with Snowflake focused on building multimodal data pipelines that convert images and audio into structured text via OCR and ASR, generate timestamped video descriptions using vision language models, and enable retrieval across slides, audio, and video with a multimodal RAG pipeline (source: DeepLearning.AI). As reported by DeepLearning.AI, the course, taught by Gilberto Hernandez, targets practitioners who need production-grade pipelines for unstructured enterprise data, highlighting concrete workflows for indexing, feature extraction, and cross-modal search that can reduce manual tagging costs and accelerate knowledge discovery in modern data stacks (source: DeepLearning.AI). According to DeepLearning.AI, the Snowflake collaboration signals growing enterprise demand for native multimodal data capabilities, creating opportunities for data teams to standardize OCR/ASR processing, integrate VLM-based video understanding, and operationalize multimodal retrieval for analytics and compliance use cases (source: DeepLearning.AI).

According to @satyanadella, pairing geotechnical data with AI is helping New Zealand build better infrastructure; as reported by Microsoft Source Asia, New Zealand agencies and engineering partners are using Azure AI to integrate borehole logs, lidar, and seismic datasets to accelerate site characterization, reduce ground risk, and cut design time for roads and utilities. According to Microsoft Source Asia, AI models on Azure ingest unstructured PDFs and legacy logs with OCR and vector search, then generate geotechnical summaries and ground condition predictions that inform foundation choices and slope stability analyses. As reported by Microsoft Source Asia, this approach improves data discoverability across councils, enables scenario testing for extreme weather resilience, and shortens consent and tender cycles for contractors, creating cost and schedule certainty. According to Microsoft Source Asia, the initiative also standardizes data governance and privacy on Microsoft Cloud, enabling cross-project reuse of subsurface knowledge while meeting public-sector compliance requirements.

According to The Rundown AI, Anthropic has launched Glasswing, accessible via anthropic.com/glasswing. According to Anthropic’s announcement, Glasswing is a new multimodal vision model designed to interpret complex images, documents, and UI screenshots with improved grounding and reasoning, positioning it for enterprise workflows in compliance, analytics, and agentic automation. As reported by Anthropic, Glasswing integrates with Claude and API tool use, enabling retrieval-augmented visual QA, structured extraction from PDFs, and step-by-step visual reasoning, which can reduce manual review time and enhance data accuracy in document-heavy sectors such as finance and healthcare. According to Anthropic, early benchmarks highlight stronger performance on chart understanding, OCR robustness, and multi-turn visual dialogues compared to prior Claude Vision releases, signaling competitive pressure on OpenAI and Google in multimodal enterprise use cases. As reported by The Rundown AI, the release page provides product details and developer resources, indicating near-term opportunities for SaaS vendors to add visual copilot features, automated reporting, and UI-testing agents powered by Glasswing.

According to DeepLearning.AI on Twitter, many of the highest ROI AI deployments focus on back‑office workflows—invoice processing, document information extraction, data integration, and day‑to‑day reliability—rather than chatbots. As reported by DeepLearning.AI, it published a four‑part learning path covering: Document AI from OCR to agentic document extraction, preprocessing unstructured data for LLM applications, functions tools and agents with LangChain, and improving accuracy of LLM applications. According to DeepLearning.AI, these resources target production use cases like automated invoicing and document pipelines, offering step‑by‑step guidance on OCR selection, schema design, retrieval, tool use, and evaluation that can reduce manual processing costs and improve data quality in enterprise systems.

According to @godofprompt, the linked paper introduces an arXiv study on a vision‑language model that targets ultra‑high‑resolution inputs. As reported by arXiv, the model processes 4K images end‑to‑end and improves zero‑shot OCR, chart understanding, and document QA without task‑specific fine‑tuning. According to the paper, benchmarking shows competitive results on DocVQA and ChartQA while maintaining robust general VLM reasoning. As noted by the authors on arXiv, the approach uses tiled feature aggregation and resolution‑aware positional encoding to preserve small‑object details at scale. For businesses, this enables automated document intake, invoice parsing, and retail shelf analytics from native‑resolution imagery, according to the arXiv evaluation and use‑case discussion.

According to DeepLearningAI on Twitter, a new course in collaboration with LandingAI titled 'Document AI: From OCR to Agentic Doc Extraction' is being launched to help users automate the process of extracting and reformatting data from documents. The course promises to teach participants how to use advanced OCR and AI-driven document extraction tools, which can significantly reduce manual data entry and streamline business workflows. As reported by DeepLearningAI, this education initiative targets professionals seeking to leverage document AI for enhanced productivity and operational efficiency.

According to DeepLearning.AI, their new course with LandingAI, 'Document AI: From OCR to Agentic Doc Extraction,' teaches users to extract information from complex documents, including those with handwritten formulas, nested captions, and overlapping watermarks. The curriculum covers practical applications of optical character recognition, layout detection, and advanced document reading, offering professionals actionable skills for automating data extraction in business workflows. As reported by DeepLearning.AI on Twitter, this course addresses growing industry needs for intelligent, agent-driven document processing.

According to Andrew Ng (@AndrewYNg), LandingAI has launched a new course titled 'Document AI: From OCR to Agentic Doc Extraction,' taught by David Park and Andrea Kropp (source: Andrew Ng on Twitter, Jan 14, 2026). The course addresses the widespread challenge of extracting structured data from unstructured documents such as PDFs and JPEGs. It covers practical techniques for building agentic document extraction systems using advanced optical character recognition (OCR) and AI-driven automation. This initiative offers concrete business opportunities for enterprises dealing with large volumes of document-based data, helping them automate workflows, improve data accuracy, and enable faster decision-making through AI-powered document processing (source: Andrew Ng on Twitter, Jan 14, 2026).