July 2025 Summaries

"People Who Ship" is a video and blog series created by Senior AI Developer Advocate Apoorva Joshi at MongoDB, aimed at offering developers insights into building and deploying production-grade AI applications using MongoDB. The series features monthly interviews, such as a discussion with Noam Rubin from Vanta's AI team, who shares strategies for transitioning generative AI applications from prototype to production. Key insights cover the importance of adaptable teams, rapid prototyping with gen AI tools like Cursor, and leveraging motivated testers for feedback. Concurrently, a leadership transition at MongoDB sees Dev Ittycheria retiring as CEO, with Chirantan “CJ” Desai taking over, bringing experience from ServiceNow and Cloudflare to guide MongoDB through its next growth phase. This transition underscores MongoDB's strategic focus on utilizing AI and data-intensive applications, with Dev expressing confidence in the company's future and commitment to facilitating a smooth leadership change while planning to remain involved on the Board.

In an evolving technological landscape, organizations must continuously adapt to maintain competitiveness and efficiency. AppMap, founded in 2020, partners with MongoDB Atlas to help developers optimize MongoDB deployments by offering AI-driven insights and interactive diagrams that enhance application performance and streamline troubleshooting. This collaboration aims to reduce development cycles and costs by simplifying the understanding of complex architectures and providing personalized recommendations for query optimization and database interactions. MongoDB Atlas complements this with its robust cloud database services, supporting scalable and secure data management. Additionally, the text discusses the transition in MongoDB's leadership, with Dev Ittycheria announcing his retirement as CEO and the appointment of Chirantan "CJ" Desai as his successor. Desai brings extensive growth-at-scale experience from his tenure at ServiceNow and Cloudflare, positioning MongoDB for its next phase of growth amidst the rise of AI and data-intensive applications. The leadership change is part of MongoDB's strategic evolution to harness new opportunities and drive innovation, while maintaining a focus on delivering customer value and enhancing developer productivity through integrated platforms like Crossplane and Kubernetes.

As companies transition from traditional DevOps to Internal Developer Platforms (IDPs), they aim to reduce the cognitive load on developers by centralizing tooling and governance, often managed by platform engineering teams. This approach retains the self-service benefits of DevOps while alleviating the burden on developers, allowing them to focus on application development without needing to understand the intricacies of infrastructure management. MongoDB's Atlas Kubernetes Operator facilitates this shift by enabling the management of Atlas resources declaratively through Kubernetes-native workflows, integrating seamlessly with systems like Crossplane. Crossplane further enhances this by enabling infrastructure management across environments using standard Kubernetes APIs, allowing for consistent, declarative, and version-controlled infrastructure provisioning. By abstracting infrastructure provisioning behind Compositions and Composite Resources, Crossplane allows platform engineers to define reusable blueprints for common services, simplifying decision-making for developers and facilitating changes in centrally managed templates. This method empowers developers to deliver faster while maintaining security and governance oversight, thereby meeting business requirements efficiently.

As modern applications increase in complexity, seamless data integration is essential, and MongoDB Atlas's unified database platform is pivotal for developers aiming to create powerful applications. The Delbridge Data API, now generally available, offers a modern solution for organizations to extract more value from MongoDB systems, focusing on scalability, security, and customization. It simplifies development by enabling frontend applications to access data directly, reducing the need for custom backend infrastructure. This is particularly useful for initial projects and prototypes, but as applications expand, more sophisticated solutions are required to manage complexity, such as integrating custom business logic and ensuring compliance with regulatory standards. The Delbridge API provides a streamlined, developer-friendly alternative to the deprecated MongoDB Data API, enhancing functionality to meet modern application demands while allowing teams greater control over data requests and security. It aligns with evolving business practices like microservices architecture and hybrid cloud strategies, acting as a customizable gateway between MongoDB and applications. A real-world example is a ride-sharing platform that uses the Delbridge Data API to optimize driver assignments and pricing dynamically, improving efficiency and reliability. The API's customization features, scalability, and enhanced observability empower businesses to grow strategically and meet compliance needs, offering a smooth migration process that minimizes disruption. MongoDB also announces a leadership transition, with Dev Ittycheria retiring as CEO and Chirantan “CJ” Desai stepping in to lead MongoDB into its next phase of growth. This transition comes as MongoDB is well-positioned to capitalize on the rise of AI and data-intensive applications, with a solid strategy and leadership team in place to capture forthcoming innovation opportunities.

Voyage-context-3 is a newly introduced contextualized chunk embedding model designed to improve retrieval accuracy by capturing both local chunk content and broader document context without requiring manual metadata or context augmentation. It outperforms existing models like OpenAI-v3-large and Cohere-v4 by significant margins in both chunk-level and document-level retrieval tasks, while being simpler, faster, and more cost-effective. The model supports various dimensions and quantization options thanks to Matryoshka learning and quantization-aware training, dramatically reducing vector database storage costs while maintaining high retrieval quality. By intelligently incorporating document-level context into chunk embeddings, voyage-context-3 enhances retrieval performance and reduces sensitivity to chunking strategies, making it an efficient drop-in replacement for standard context-agnostic embeddings in retrieval-augmented generation (RAG) pipelines.

Retrieval-augmented generation (RAG) is revolutionizing AI applications by grounding generated responses in factual data, reducing hallucinations, and improving precision and contextual relevance. This comprehensive guide delves into deploying a production-ready RAG application using MongoDB Atlas and Cohere Command R+, expanding on the official Cohere and MongoDB RAG documentation. It details building a complete RAG pipeline, focusing on data flow, retrieval, and generation, and enhancing answer quality through reranking and flexible deployment with Docker Compose. The integration of MongoDB Atlas as a vector store and chat memory, combined with Cohere Command R+, offers a powerful approach for creating scalable, high-performance systems for grounded generative AI. This synergy enables applications to deliver fast, accurate, and contextually informed responses by leveraging real-world data, thus representing a compelling method for developing next-generation AI applications.

Retailers are facing the challenge of meeting rising customer expectations in an omnichannel world, where seamless and consistent experiences across all touchpoints are crucial. Store associates play a vital role in bridging online and offline channels, but their effectiveness is often hampered by siloed systems and the lack of real-time inventory data. To address these issues, a unified commerce approach is recommended, integrating all sales channels and data systems into a seamless platform. MongoDB Atlas is highlighted as a solution that provides a modern, flexible database capable of securely integrating complex data from multiple systems, offering a 360-degree view of the business. This unified data platform empowers store associates by providing them with real-time, accurate information, boosting their ability to deliver personalized and efficient customer service, ultimately leading to increased sales and customer loyalty.

As of September 26, 2025, MongoDB is shifting the focus of its AI Applications Program (MAAP) towards developing strategic partnerships, particularly in collaboration with IBM’s Watsonx.ai and MongoDB Atlas to deliver AI-driven solutions for financial institutions. This integration aims to create an intelligent finance assistant capable of providing personalized, real-time financial insights by combining MongoDB Atlas's advanced vector search capabilities with IBM Watsonx.ai's generative AI models. The architecture facilitates natural language understanding, data retrieval, and intelligent response generation, addressing the limitations of static systems and generic chatbots. MongoDB Atlas is chosen for its robust data management and hybrid search capabilities, while IBM Watsonx.ai offers enterprise-grade AI models suited for the finance sector. This partnership underlines the potential of combining vector search and AI to enhance user experiences and operational efficiency in the financial industry, and it represents a strategic move by MongoDB to leverage AI technologies for future growth.

Google's Datastream service has introduced public preview support for MongoDB as a source, enabling seamless data ingestion from MongoDB databases into Google's BigQuery and Cloud Storage for real-time analytics and decision-making. MongoDB Atlas is renowned for its flexibility, scalability, and performance, making it an ideal choice for applications needing agile schema evolution and handling varied data types. By integrating MongoDB data with BigQuery, users can unlock advanced analytics, machine learning, and AI capabilities, driving new insights and business growth. Datastream facilitates real-time data replication with low latency and high reliability, eliminating complex batch processes and reducing operational overhead. The announcement also covers MongoDB's role in leveraging AI, highlighting customer success stories, such as Ubuy, Financial Times, CentralReach, and Base39, which have used MongoDB Atlas to enhance their applications with AI-driven insights. Additionally, the text discusses MongoDB's leadership transition, with Dev Ittycheria retiring as CEO and Chirantan "CJ" Desai taking over, emphasizing the company's strategic alignment for future growth in the AI and data-driven landscape.

In a detailed exploration of MongoDB's impact across various industries, the text highlights the transformative role of MongoDB Atlas in leveraging AI for innovation and efficiency. Several companies are showcased, including Ubuy, which improved search performance and user engagement by migrating to MongoDB Atlas, and the Financial Times, which enhanced content discovery through AI-powered hybrid search. CentralReach utilized MongoDB's capabilities to streamline autism care, while Base39 revolutionized credit analysis with AI-driven insights. Additionally, PLAID optimized real-time data processing and reduced costs using MongoDB Atlas Stream Processing. The document also announces a leadership transition at MongoDB, with Dev Ittycheria retiring as CEO and Chirantan “CJ” Desai taking over, emphasizing CJ's experience in scaling companies and his alignment with MongoDB's strategic vision. Dev reflects on his tenure and expresses confidence in MongoDB's future under CJ's leadership, underscoring the company's readiness to capitalize on the rise of AI and data-intensive applications.

PLAID, Inc., a Tokyo-based company that has been a MongoDB customer since 2015, encountered challenges in scaling its real-time data processing capabilities when migrating from a self-hosted MongoDB instance to MongoDB Atlas. Initially using Kafka connectors to stream data to Google BigQuery, the company faced complexities and rising costs due to the number of pipelines required. To address these issues, PLAID implemented MongoDB Atlas Stream Processing, which offered an integrated and cost-effective solution for real-time data processing. This transition allowed PLAID to replace some Kafka connectors, streamline connection management, and optimize costs by leveraging Stream Processing Instances (SPIs) with a more predictable pricing structure. The implementation resulted in significant cost savings, improved scalability, and enhanced real-time data capabilities, enabling PLAID to efficiently process over 3 million events per day while maintaining operational stability and simplifying management. The success with Atlas Stream Processing underscores its value for organizations aiming to streamline data integration pipelines and effectively leverage real-time data.

In this detailed analysis, Andrew Morgan discusses a performance issue encountered in a customer's application using MongoDB, which was traced back to inefficient indexing of embedded objects. The application suffered from slow read performance due to 15 indexes consuming more space than available RAM, leading to slow disk fetches. The text emphasizes the importance of using compound and partial indexes effectively, avoiding indexing entire objects, and ensuring queries match the indexed fields to optimize performance. The document also highlights the utility of MongoDB design reviews, which can help applications meet performance requirements by advising on appropriate schema and indexing strategies. Additionally, the article underscores the significance of a well-informed indexing approach to maintain database efficiency even as data volumes grow.

In a rapidly changing geopolitical climate, the global automotive industry faces significant disruptions, particularly due to the reintroduction of tariffs, which have impacted production cycles and model-year transitions. This has resulted in a notable decrease in new-model vehicle availability and overall inventory, pressuring consumer pricing and inventory management. Traditionally, inventory classification relied on ABC analysis, which segments items by value, but this method is criticized for its limited criteria. A more comprehensive multi-criteria inventory classification (MCIC) approach, incorporating factors like lead-time and durability, is proposed. However, the importance of unstructured data, such as customer feedback and product reviews, is increasingly recognized. Using large language models (LLMs), insights from these unstructured data sources can be vectorized to enhance inventory classification models, shifting from reactive to predictive management. MongoDB facilitates this by enabling AI-driven inventory classification through vector embeddings and dynamic criteria generation, leveraging structured and unstructured data. This approach aims to provide a more nuanced understanding of product value and demand, adapting to the evolving needs of the automotive industry.

MongoDB has introduced a Multimodal Search Python Library to facilitate the development of applications leveraging diverse data types such as text, images, and complex documents, addressing challenges developers face in searching and retrieving information across these formats. The library integrates MongoDB Atlas Vector Search, AWS S3, and Voyage AI's multimodal embedding model, voyage-multimodal-3, to streamline the processing, storage, and indexing of data, enhancing information retrieval and user experiences. Additionally, MongoDB has announced a leadership transition, with Dev Ittycheria retiring as CEO in November 2025, to be succeeded by Chirantan "CJ" Desai, who brings extensive experience from ServiceNow and Cloudflare. This transition aims to propel MongoDB's next phase of growth, capitalizing on opportunities in AI and data-intensive applications, while Ittycheria will remain on the Board to ensure a smooth transition.

The latest cohort of MongoDB Community Champions has been announced, featuring a diverse group of 47 members, including 21 newcomers from various countries, who serve as key liaisons between MongoDB and its global community. These champions, who hold roles ranging from engineering leads to chief architects, share their expertise through media and events, contributing to product development and community engagement. The program offers its members unique opportunities such as access to executives, product roadmaps, and an annual summit, enhancing their professional status in the tech community. Additionally, the blog highlights the Matryoshka Representation Learning (MRL) approach, which creates flexible, multi-fidelity embeddings, allowing for efficient data processing and comparison by adapting vector dimensions to specific needs. MRL, combined with quantization techniques, enables Voyage AI models to offer scalable, efficient data retrieval solutions with MongoDB Atlas Vector Search, balancing accuracy, storage, and performance.

In an exploration of safety in industrial operations, the blog post by Humza Akhtar delves into the challenges faced by heavy-asset industries, such as aerospace, shipbuilding, and construction, where tight environments, time pressures, and production targets often lead to unsafe shortcuts. The European Maritime Safety Agency's data, which attributes 80% of marine incidents to human factors, underscores the need for a proactive approach to safety rather than reactive measures. The blog suggests utilizing MongoDB as a unified operational data store to consolidate sensor telemetry, worker decisions, and contextual factors, enabling organizations to generate proactive insights and improve safety policies. By integrating game theory and MongoDB’s flexible document model, the post advocates for modeling and analyzing worker decisions as strategic interactions, allowing the simulation of safety conditions and the development of informed policies. The use of MongoDB’s features like time series data storage, game-theoretic decision modeling, and risk scoring aims to create a behavior-aware safety simulation engine that anticipates and mitigates risky behavior. The approach emphasizes the importance of real-time data integration and adaptable policy design to enhance safety in industrial environments.

Harshad Dhavale's article discusses the advancements in embedding models, particularly focusing on Matryoshka Representation Learning (MRL), a novel approach that addresses the limitations of traditional fixed-size embedding models by allowing flexible, multi-fidelity embeddings. These models, which convert unstructured data into numerical vectors, are crucial for applications like semantic search and recommendation systems. Traditional embeddings often face challenges such as inflexibility, high computational load, and information loss when truncated. MRL, inspired by Russian nesting dolls, enables a single model to produce embeddings that can be truncated to various dimensions without losing semantic quality. The training process for MRL involves computing multiple loss values for different truncated prefixes, incentivizing the model to pack crucial information into the earliest dimensions, thereby retaining accuracy with fewer dimensions. MRL contrasts with quantization, which reduces embedding size by compressing precision, as MRL focuses on dimensional flexibility. Voyage AI exemplifies the use of MRL by combining it with quantization for ultimate efficiency, allowing dynamic choices between space, latency, and quality, leading to efficient retrieval and reduced infrastructure costs.

In a detailed analysis of AI agent architectures, the discussion centers on the contrasting approaches of Anthropic and Cognition in building intelligent systems, emphasizing the critical role of memory management. Anthropic advocates for a multi-agent system, suitable for extensive research tasks that require sophisticated memory techniques like compression and external storage to manage complex, distributed memory across different agents. Conversely, Cognition supports a single-agent design, which excels in tasks needing consistent decision-making such as conversational AI, emphasizing context engineering to maintain memory flow. Both approaches underscore the necessity of robust memory systems for agent reliability and capability, highlighting that the choice between multi-agent and single-agent systems largely depends on the specific application mode, whether it be research, conversation, or coding. The text suggests that as AI evolves, memory management will become a pivotal aspect of AI engineering, requiring specialized skills to create systems that can effectively remember, reason, and adapt over time.

The text highlights the challenges and transformations in data infrastructure and database management necessary to harness the full potential of AI. Traditional databases, designed for structured data, often become bottlenecks due to their rigid schemas and separation of operational and analytical systems. The piece underscores the need for a modern data architecture that embraces flexible schemas, integrates operational and analytical data, and utilizes databases that align with developers' needs, such as MongoDB Atlas on Azure. The retail and healthcare sectors are highlighted as examples where modern databases have enabled real-time data-driven decisions and streamlined processes. The shift from using databases merely as storage to strategic enablers of intelligent applications is emphasized, with benefits including enhanced developer productivity and reduced complexity. The text also discusses the inefficiencies and costs associated with traditional relational databases due to their complexity and the need for significant computational resources, contrasting this with MongoDB's more efficient document model. Furthermore, it offers guidance on transitioning to a modern data architecture, cautioning against common pitfalls like simultaneous modernization of all systems and creating new data silos. Finally, the text includes an announcement from MongoDB's CEO, Dev Ittycheria, about his upcoming retirement and the appointment of Chirantan "CJ" Desai as his successor, highlighting the strategic leadership shift as MongoDB prepares for its next phase of growth amid the rise of AI and data-intensive applications.

Security operations teams are increasingly challenged by the complexity of cloud-native applications and the resulting flood of logs and events, which traditional security tools often fail to efficiently analyze due to their isolated approach. Graph analytics, exemplified by the PuppyGraph tool, offers a solution by modeling security data as interconnected nodes, allowing real-time analysis of AWS CloudTrail data within MongoDB without the need for data movement. By utilizing MongoDB's flexible schema for high-throughput ingestion of unstructured security logs, PuppyGraph enhances threat detection by revealing complex relationships and attack patterns that static alerts may miss. Through the integration of MongoDB and PuppyGraph, teams can conduct sophisticated investigations into privilege escalation, user behavior, and access patterns, transforming log collections into interactive graphs that provide a comprehensive view of security incidents as they unfold. This approach not only makes security data more accessible and interpretable but also allows for a seamless and scalable implementation of graph analytics without altering existing infrastructure.

Relational databases, grounded in the principle of normalization, aim to minimize redundancy by distributing data across multiple interconnected tables, which complicates data structures and increases costs for developers, infrastructure administrators, and portfolio managers. Developers face challenges navigating and joining these tables, making relational databases less efficient and more time-consuming compared to MongoDB's document model, which consolidates data into single objects, reducing complexity and enhancing productivity. Infrastructure administrators incur higher computational costs due to the resource-intensive nature of JOIN operations required by normalized databases, necessitating additional hardware or cloud resources, unlike MongoDB's efficient design that supports data-intensive workloads on existing infrastructure. Portfolio managers overseeing application suites find relational databases slow to deliver features due to their complexity and the larger teams required for database management, whereas MongoDB's flexibility and efficiency contribute to faster feature delivery and lower costs. The overarching message emphasizes MongoDB's advantages over traditional relational databases, particularly in terms of reducing complexity, costs, and time to market, while enhancing agility and efficiency, making it a preferred choice for modern data management needs.

MongoDB has introduced External Functions in its Atlas Stream Processing, enabling direct invocation of AWS Lambda from streaming pipelines to enrich, validate, and transform data in real-time. This feature, available through the $externalFunction pipeline stage, integrates external logic services like AWS Lambda, allowing users to leverage existing business logic and perform AI/ML inference without embedding the logic within the pipeline itself. AWS Lambda, a scalable serverless compute service, can be used synchronously or asynchronously within the pipeline depending on whether an immediate response is required. A practical application of this is demonstrated through a solar energy company that uses synchronous external functions to monitor real-time telemetry from solar devices, enriching data with status updates and diagnostics before storing it in MongoDB Atlas. This capability enhances stream processing by making data processing smarter and more adaptable, with potential use cases in various fields like fraud detection and personalization.

In 2019, MongoDB's replication team embarked on developing a new, safe reconfiguration protocol for their database systems, addressing the known bugs and limitations of the existing legacy protocol. This new protocol, designed to be logless and to minimize changes to the existing gossip-based system, had to ensure high availability and fault tolerance when dynamically reconfiguring replica sets. By utilizing formal specification and model checking tools like TLA+ and TLC, the team rapidly iterated on their design, ensuring rigorous correctness while simplifying the protocol's implementation. The resulting protocol demonstrated novel performance benefits by decoupling reconfiguration from the main database operation log and enhanced reliability by eliminating the need for dual protocol maintenance. Since its implementation in MongoDB 4.4, the protocol has proven robust and reliable, serving as a foundation for additional features while avoiding major bugs, thus underscoring the value of formal methods in protocol design and optimization.

MongoDB has introduced Query Shape Insights, a new feature within MongoDB Atlas designed to offer a comprehensive view of query performance across clusters, specifically targeting resource-intensive query shapes. This feature aggregates data on the most demanding query shapes, facilitating root cause analysis, optimizing workflows, and enhancing operational efficiency. Unlike previous tools focusing on individual queries, Query Shape Insights provides a holistic perspective, ranking the top 100 query shapes by execution time and offering detailed metrics such as operation count, documents examined, and bytes read. This allows for a more efficient identification of performance bottlenecks, particularly in microservices environments. With integration into MongoDB’s existing observability suite, users can quickly transition from detection to resolution, supported by features that include interactive analysis tools, flexible filtering options, and programmatic access through MongoDB’s Admin API. Additionally, the feature adapts based on cluster tier, supporting various workload sizes and enabling trend tracking over a seven-day window. Overall, Query Shape Insights represents a significant advancement in performance monitoring, allowing MongoDB Atlas users to manage, investigate, and respond to performance issues more effectively.

As modern connected vehicles generate extensive and diverse data, the automotive industry faces challenges in efficiently processing, storing, and analyzing this information. With a single car producing up to 25GB of data per hour, the need for a flexible, reliable data solution has become critical. Organizations like the Connected Vehicle Systems Alliance (COVESA) have introduced standards such as the Vehicle Signal Specification (VSS) to improve interoperability and streamline data use. Document-oriented databases like MongoDB are increasingly preferred over relational databases to handle the complexity and scale of vehicle data, offering benefits such as reduced complexity, scalability by design, and adaptability to changing vehicle platforms. MongoDB's architecture supports AI workloads and is particularly effective for real-time applications like fleet management, as demonstrated by companies like Volvo Connect and SHARE NOW. The blog also highlights MongoDB's recent advancements, including a prototype for modeling vehicle signal data and an integration with LangChain for natural language querying, which further enhances its capacity to meet modern data demands in the automotive sector.

Building event-driven architectures (EDAs) can be challenging, especially when integrating complex cloud components with local development services. A unique development workflow showcased by a demo application illustrates how to balance local service integration with cloud stream processing using MongoDB Atlas Stream Processing and ngrok. This approach allows developers to work directly from their local environment, providing convenience and efficiency by securely interacting with cloud services. MongoDB Atlas Stream Processing simplifies the development of EDAs by allowing processing logic to be defined directly within MongoDB Atlas, using familiar syntax and managed infrastructure. The demo application simulates a real-time order fulfillment process, demonstrating the orchestration of event-driven architecture through MongoDB Atlas Stream Processing. It features services like shopping cart and order processing, with events ingested from MongoDB or Apache Kafka and enriched by processors that trigger actions in different services. This setup illustrates the ease of building and testing sophisticated, event-driven applications locally, highlighting MongoDB Atlas Stream Processing's capability to lower the barriers to entry for real-time EDAs by integrating seamlessly with local development environments.