Android Earthquake Alerts: A global system for early warning (opens in new tab)

Google’s Android Earthquake Alerts system utilizes the onboard accelerometers of billions of smartphones to create a global, crowdsourced seismic network. By detecting the initial P-waves of an earthquake and rapidly processing aggregate data, the system provides critical early warnings to regions that often lack traditional, expensive seismic infrastructure. This technological shift has expanded earthquake early warning access from roughly 250 million people to over 2.5 billion worldwide. ### Leveraging On-Device Accelerometers * Mobile accelerometers, typically used for screen orientation, function as mini-seismometers capable of detecting the initial, fast-moving P-waves of an earthquake. * When a stationary phone detects these vibrations, it sends a signal along with a coarse location to a centralized detection server. * The system aggregates these signals to confirm the event and estimate its magnitude before the slower, more destructive S-waves reach the population. ### Global Reach and Implementation * Active in nearly 100 countries, the system has detected over 18,000 earthquakes ranging from M1.9 to M7.8. * The system has issued alerts for over 2,000 significant earthquakes, resulting in approximately 790 million alerts sent to users globally. * By utilizing existing consumer hardware, the system serves as a "global safety net" for earthquake-prone regions that cannot afford traditional ground-based sensor networks. ### Magnitude Estimation and Accuracy * A primary technical challenge is the trade-off between speed and accuracy; the first few seconds of data are limited but essential for providing early warnings. * Over three years, the system's median absolute error for initial magnitude estimates has been reduced from 0.50 to 0.25. * The accuracy of these smartphone-based detections is now comparable to—and in some cases exceeds—the performance of established traditional seismic networks. ### User Alerts and Real-World Impact * The system delivers two tiers of notifications: "BeAware" alerts for light shaking and "TakeAction" alerts, which use full-screen takeovers and loud sounds for intense shaking. * During a magnitude 6.7 earthquake in the Philippines, the system issued alerts 18.3 seconds after the quake began, providing users further from the epicenter up to 60 seconds of lead time. * To maintain privacy, the system relies on coarse location data and requires users to have Wi-Fi or cellular connectivity and location settings enabled. For users in seismic zones, ensuring that Android Earthquake Alerts and location services are enabled provides a vital layer of protection. This crowdsourced model demonstrates how ubiquitous consumer technology can be repurposed to provide essential public safety infrastructure on a planetary scale.

LY's Tech Conference, 'Tech (opens in new tab)

LY Corporation’s Tech-Verse 2025 conference highlighted the company's strategic pivot toward becoming an AI-centric organization through the "Catalyst One Platform" initiative. By integrating the disparate infrastructures of LINE and Yahoo! JAPAN into a unified private cloud, the company aims to achieve massive cost efficiencies while accelerating the deployment of AI agents across its entire service ecosystem. This transformation focuses on empowering engineers with AI-driven development tools to foster rapid innovation and deliver a seamless, "WOW" experience for global users. ### Infrastructure Integration and the Catalyst One Platform To address the redundancies following the merger of LINE and Yahoo! JAPAN, LY Corporation is consolidating its technical foundations into a single internal ecosystem known as the Catalyst One Platform. * **Private Cloud Advantage:** The company maintains its own private cloud to achieve a four-fold cost reduction compared to public cloud alternatives, managed by a lean team of 700 people supporting 500,000 servers. * **Unified Architecture:** The integration spans several layers, including Infrastructure (Project "DC-Hub"), Cloud (Project "Flava"), and specialized Data and AI platforms. * **Next-Generation Cloud "Flava":** This platform integrates existing services to enhance VM specifications, VPC networking, and high-performance object storage (Ceph and Dragon). * **Information Security:** A dedicated "SafeOps" framework is being implemented to provide governance and security across all integrated services, ensuring a safer environment for user data. ### AI Strategy and Service Agentization A core pillar of LY’s strategy is the "AI Agentization" of all its services, moving beyond simple features to proactive, personalized assistance. * **Scaling GenAI:** Generative AI has already been integrated into 44 different services within the group. * **Personalized Agents:** The company is developing the capacity to generate millions of specialized agents that can be linked together to support the unique needs of individual users. * **Agent Ecosystem:** The goal is to move from a standard platform model to one where every user interaction is mediated by an intelligent agent. ### AI-Driven Development Transformation Beyond user-facing services, LY is fundamentally changing how its engineers work by deploying internal AI development solutions to all staff starting in July. * **Code and Test Automation:** Proof of Concept (PoC) results showed a 96% accuracy rate for "Code Assist" and a 97% reduction in time for "Auto Test" procedures. * **RAG Integration:** The system utilizes Retrieval-Augmented Generation (RAG) to leverage internal company knowledge and guidelines, ensuring high-quality, context-aware development support. * **Efficiency Gains:** By automating repetitive tasks, the company intends for engineers to shift their focus from maintenance to creative service improvement and innovation. The successful integration of these platforms and the aggressive adoption of AI-driven development tools suggest that LY Corporation is positioning itself to be a leader in the "AI-agent" era. For technical organizations, LY's model serves as a case study in how large-scale mergers can leverage private cloud infrastructure to fund and accelerate a company-wide AI transition.

Code Quality Improvement Techniques Part (opens in new tab)

The builder pattern is frequently overused in modern development, often leading to code that is less robust than it appears. While it provides a fluent API, it frequently moves the detection of missing mandatory fields from compile-time to runtime, creating a "house of sand" that can collapse unexpectedly. By prioritizing constructors and factory functions, developers can leverage the compiler to ensure data integrity and build more stable applications. ### Limitations of the Standard Builder Pattern * In a typical builder implementation, mandatory fields are often initialized as nullable types and checked for nullity only when the `.build()` method is called. * This reliance on runtime checks like `checkNotNull` means that a developer might forget to set a required property, leading to an `IllegalStateException` during execution rather than a compiler error. * Unless the platform or a specific library (like an ORM) requires it, the boilerplate of a builder often hides these structural weaknesses without providing significant benefits. ### Strengthening Foundations with Constructors and Defaults * Using a class constructor or a factory function is often the simplest and most effective way to prevent bugs related to missing data. * In languages like Kotlin, the need for builders is further reduced by the availability of default parameters and named arguments, allowing for concise instantiation even with many optional fields. * If a builder must be used, mandatory arguments should be required in the builder's own constructor (e.g., `Builder(userName, emailAddress)`) to ensure the object is never in an invalid state. ### Managing Creation State and Pipelines * Developers sometimes pass a builder as an "out parameter" to other functions to populate data, which can obscure the flow of data and reduce readability. * A better approach is to use functions that return specific values, which are then passed into a final constructor, keeping the logic functional and transparent. * For complex, multi-stage creation logic, defining distinct types for each stage—such as moving from a `UserAccountModel` to a `UserProfileViewComponent`—can ensure that only valid, fully-formed data moves through the pipeline. ### Appropriate Use of Terminal Operations * The builder-like syntax is highly effective when implementing "terminal operations," where various transformations are applied in an arbitrary order before a final execution. * This pattern is particularly useful in image processing or UI styling (e.g., `.crop().fitIn().colorFilter()`), where it serves as a more readable alternative to deeply nested decorator patterns. * In these specific cases, the pattern facilitates a clear sequence of operations while maintaining a "last step" (like `.createBitmap()`) that signals the end of the configuration phase. Prioritize the use of constructors and factory functions to catch as many errors as possible during compilation. Reserve the builder pattern for scenarios involving complex terminal operations or when dealing with restrictive library requirements that demand a specific instantiation style.

Graph foundation models for relational data (opens in new tab)

Google researchers have introduced Graph Foundation Models (GFMs) as a solution to the limitations of traditional tabular machine learning, which often ignores the rich connectivity of relational databases. By representing tables as interconnected graphs where rows are nodes and foreign keys are edges, this approach enables a single model to generalize across entirely different schemas and feature sets. This shift allows for transferable graph representations that can perform inference on unseen tasks without the costly need for domain-specific retraining. ### Transforming Relational Schemas into Graphs The core methodology involves a scalable data preparation step that converts standard relational database structures into a single heterogeneous graph. This process preserves the underlying logic of the data while making it compatible with graph-based learning: * **Node Mapping:** Each unique table is treated as a node type, and every individual row within that table is converted into a specific node. * **Edge Creation:** Foreign key relationships are transformed into typed edges that connect nodes across different tables. * **Feature Integration:** Standard columns containing numerical or categorical data are converted into node features, while temporal data can be preserved as features on either nodes or edges. ### Overcoming the Generalization Gap A primary hurdle in developing GFMs is the lack of a universal tokenization method, unlike the word pieces used in language models or patches used in vision models. Traditional Graph Neural Networks (GNNs) are typically locked to the specific graph they were trained on, but GFMs solve this through several technical innovations: * **Schema Agnosticism:** The model avoids hard-coded embedding tables for specific node types, allowing it to interpret database schemas it has never encountered during training. * **Feature Interaction Learning:** Instead of training on "absolute" features (like specific price distributions), the model captures how different features interact with one another across diverse tasks. * **Generalizable Encoders:** The architecture uses transferable methods to derive fixed-size representations for nodes, whether they contain three continuous float features or dozens of categorical values. ### Scaling and Real-World Application To handle the requirements of enterprise-level data, the GFM framework is built to operate on a massive scale using Google’s specialized infrastructure: * **Massive Throughput:** The system utilizes JAX and TPU infrastructure to process graphs containing billions of nodes and edges. * **Internal Validation:** The model has been tested on complex internal Google tasks, such as spam detection in advertisements, which requires analyzing dozens of interconnected relational tables simultaneously. * **Performance Benefits:** By considering the connections between rows—a factor traditional tabular baselines like decision trees often ignore—the GFM provides superior downstream performance in high-stakes prediction services. Transitioning from domain-specific models to Graph Foundation Models allows organizations to leverage relational data more holistically. By focusing on the connectivity of data rather than just isolated table features, GFMs provide a path toward a single, generalist model capable of handling diverse enterprise tasks.

Code Quality Improvement Techniques Part (opens in new tab)

The Null Object Pattern is a design technique that replaces null values with objects representing "empty" or "invalid" states to simplify code and provide functional fallbacks. While it effectively streamlines logic for collections and general data flows, using it when error conditions must be explicitly distinguished can lead to hidden bugs and reduced type safety. Developers should generally prefer statically verified types, such as Optionals or language-native nullable types, unless the error case can be seamlessly integrated into the happy-path logic. ### Benefits of the Null Object Pattern * **Code Simplification:** By returning an empty list or a "null object" instead of a literal `null`, callers can avoid repetitive null-check boilerplate. * **Functional Continuity:** It allows for uninterrupted processing in functional chains, such as using `.asSequence().map().forEach()`, because the "empty" object still satisfies the required interface. * **Fallback Provisioning:** The pattern is useful for converting errors into safe fallback values, such as displaying an "Unknown User" profile image rather than crashing or requiring complex conditional UI logic. ### Risks of Silent Failures and Logic Errors * **Bypassing Compiler Safety:** Unlike nullable types in Kotlin or Swift, which force developers to handle the `null` case, a custom null object (e.g., `UserModel.INVALID`) allows code to compile even if the developer forgets to check the object's validity. * **Identity vs. Equivalence:** Implementing the pattern requires caution regarding how the object is compared. If a null object is checked via reference identity (`==`) but the class lacks a proper `equals` implementation, new instances with the same "empty" values may not be recognized as invalid. * **Debugging Difficulty:** When a null object is used inappropriately, the program may continue to run with dummy data. This makes bugs harder to detect compared to a runtime error or a compile-time type mismatch. ### Best Practices for Type Safeness * **Prefer Static Verification:** When boundary conditions or errors must be handled differently than the "happy path," use `Optional`, `Maybe`, or native nullable types to ensure the compiler enforces error handling. * **Criteria for Use:** Reserve the Null Object Pattern for cases where the error logic is identical to the normal logic, or when multiple "empty" candidates exist that cannot be easily resolved through static typing. * **Runtime Errors as a Tool:** In dynamic or non-nullable contexts, a runtime error is often preferable to silent execution with an invalid object, as it provides a clear signal that an unexpected state has been reached. ### Recommendation To maintain high code quality, utilize the Null Object Pattern primarily for collections and UI fallbacks. For core business logic where the presence of data is critical, rely on type-safe mechanisms that force explicit handling of missing values, thereby preventing invalid states from propagating silently through the system.

MedGemma: Our most capable open models for health AI development (opens in new tab)

Google Research has expanded its Health AI Developer Foundations (HAI-DEF) collection with the release of MedGemma and MedSigLIP, a series of open, multimodal models designed specifically for medical research and application development. These models offer a high-performance, privacy-preserving alternative to closed systems, allowing developers to maintain full control over their infrastructure while leveraging state-of-the-art medical reasoning. By providing both 4B and 27B parameter versions, the collection balances computational efficiency with complex longitudinal data interpretation, even enabling deployment on single GPUs or mobile hardware. ## MedGemma Multimodal Variants The MedGemma collection utilizes the Gemma 3 architecture to process both image and text inputs, providing robust generative capabilities for healthcare tasks. * **MedGemma 27B Multimodal:** This model is designed for complex tasks such as interpreting longitudinal electronic health records (EHR) and achieves an 87.7% score on the MedQA benchmark, performing within 3 points of DeepSeek R1 at approximately one-tenth the inference cost. * **MedGemma 4B Multimodal:** A lightweight version that scores 64.4% on MedQA, outperforming most open models under 8B parameters; it is optimized for mobile hardware and specific tasks like chest X-ray report generation. * **Clinical Accuracy:** In unblinded studies, 81% of chest X-ray reports generated by the 4B model were judged by board-certified radiologists to be sufficient for patient management, achieving a RadGraph F1 score of 30.3. * **Versatility:** The models retain general-purpose capabilities from the original Gemma base, ensuring they remain effective at instruction-following and non-English language tasks while handling specialized medical data. ## MedSigLIP Specialized Image Encoding MedSigLIP serves as the underlying vision component for the MedGemma suite, but it is also available as a standalone 400M parameter encoder for structured data tasks. * **Architecture:** Based on the Sigmoid loss for Language Image Pre-training (SigLIP) framework, it bridges the gap between medical imagery and text through a shared embedding space. * **Diverse Modalities:** The encoder was fine-tuned on a wide variety of medical data, including fundus photography, dermatology images, histopathology patches, and chest X-rays. * **Functional Use Cases:** It is specifically recommended for tasks involving classification, retrieval, and search, where structured outputs are preferred over free-text generation. * **Data Retention:** Training protocols ensured the model retained its ability to process natural images, maintaining its utility for hybrid tasks that mix medical and non-medical visual information. ## Technical Implementation and Accessibility Google has prioritized accessibility for developers by ensuring these models can run on consumer-grade or limited hardware environments. * **Hardware Compatibility:** Both the 4B and 27B models are designed to run on a single GPU, while the 4B and MedSigLIP versions are adaptable for edge computing and mobile devices. * **Open Resources:** To support the community, Google has released the technical reports, model weights on Hugging Face, and implementation code on GitHub. * **Developer Flexibility:** Because these are open models, researchers can fine-tune them on proprietary datasets without compromising data privacy or being locked into specific cloud providers. For medical AI development, the choice of model should depend on the specific output requirement: MedGemma is the optimal starting point for generative tasks like visual question answering or report drafting, while MedSigLIP is the preferred tool for building high-speed classification and image retrieval systems.

Making group conversations more accessible with sound localization (opens in new tab)

Google Research has introduced SpeechCompass, a system designed to improve mobile captioning for group conversations by integrating multi-microphone sound localization. By shifting away from complex voice-recognition models toward geometric signal processing, the system provides real-time speaker diarization and directional guidance through a color-coded visual interface. This approach significantly reduces the cognitive load for users who previously had to manually associate a wall of scrolling text with different speakers in a room. ## Limitations of Standard Mobile Transcription * Traditional automatic speech recognition (ASR) apps concatenate all speech into a single block of text, making it difficult to distinguish between different participants in a group setting. * Existing high-end solutions often require audio-visual separation, which needs a clear line of sight from a camera, or speaker embedding, which requires pre-registering unique voiceprints. * These current methods can be computationally expensive and often fail in spontaneous, mobile environments where privacy and setup speed are priorities. ## Hardware and Signal Localization * The system was prototyped in two forms: a specialized phone case featuring four microphones connected to an STM32 microcontroller and a software-only implementation for standard dual-microphone smartphones. * While dual-microphone setups are limited to 180-degree localization due to "front-back confusion," the four-microphone array enables full 360-degree sound tracking. * The system utilizes Time-Difference of Arrival (TDOA) and Generalized Cross Correlation with Phase Transform (GCC-PHAT) to estimate the angle of arrival for sound waves. * To handle indoor reverberations and noise, the team applied statistical methods like kernel density estimation to improve the precision of the localizer. ## Advantages of Waveform-Based Diarization * **Low Latency and Compute:** By avoiding heavy machine learning models and weights, the algorithm can run on low-power microcontrollers with minimal memory requirements. * **Privacy Preservation:** Unlike speaker embedding techniques, SpeechCompass does not identify unique voiceprints or require video, instead relying purely on the physical location of the sound source. * **Language Independence:** Because the system analyzes the differences between audio waveforms rather than the speech content itself, it is entirely language-agnostic and can localize non-speech sounds. * **Dynamic Reconfiguration:** The system adjusts instantly to the movement of the device, allowing users to reposition their phones without recalibrating the diarization logic. ## User Interface and Accessibility * The prototype Android application augments standard speech-to-text with directional data received via USB from the microphone array. * Transcripts are visually separated by color and accompanied by directional arrows, allowing users to quickly identify where a speaker is located in the physical space. * This visual feedback loop transforms a traditional transcript into a spatial map of the conversation, making group interactions more accessible for individuals who are deaf or hard of hearing.

How we created HOV-specific ETAs in Google Maps (opens in new tab)

Google Maps has enhanced its routing capabilities by introducing HOV-specific ETAs, addressing the significant speed differences between carpool and general lanes. This was achieved through a novel unsupervised learning approach that classifies historical trips into HOV or non-HOV categories without initial manual labels. The resulting system enables more precise travel predictions, helping users optimize their commutes and supporting the shift toward sustainable travel modes. ### Segment-Level Speed Distribution * The model analyzes trip segments within short, 15-minute time windows to identify patterns in aggregated, anonymized traffic data. * During peak traffic hours, researchers often observe a bimodal speed distribution where HOV lanes maintain significantly higher average speeds compared to general lanes. * The classification system distinguishes between "Scenario A," where the speed gap is dramatic (e.g., 65 mph vs. 25 mph), and "Scenario B," where HOV lanes are only marginally faster, ensuring accurate modeling even when benefits are minimal. * Individual trip points, including speed and observation time, are processed collectively to determine if a specific segment of a journey occurred in a restricted lane. ### Incorporating Lateral Distance and Soft Clustering * To refine accuracy beyond simple speed metrics, the model incorporates the estimated lateral distance of a vehicle from the center of the road. * While GPS data is inherently noisy, this spatial information helps identify lane-specific behaviors by mapping trip points to the known physical location of HOV lanes (e.g., the far-left lanes). * The system employs soft clustering techniques, calculating the probability of a point belonging to a specific cluster rather than using hard binary assignments, which better manages borderline data points. * Temporal clustering via a weighted median approach is used to prioritize more recent traffic observations, ensuring the model accounts for the most current road conditions and availability constraints. By integrating these segment-level classifications into full-trip analyses, Google Maps can train its ETA prediction models on high-fidelity, lane-specific data. This implementation provides users with a more realistic view of their travel options, encouraging the use of high-occupancy lanes to reduce individual travel time, urban congestion, and overall emissions.

REGEN: Empowering personalized recommendations with natural language (opens in new tab)

Google Research has introduced REGEN, a benchmark dataset designed to evolve recommender systems from simple item predictors into conversational agents capable of natural language interaction. By augmenting the Amazon Product Reviews dataset with synthetic critiques and narratives using Gemini 1.5 Flash, the researchers provide a framework for training models to understand user feedback and explain their suggestions. The study demonstrates that integrating natural language critiques significantly improves recommendation accuracy while enabling models to generate personalized, context-aware content. ### Composition of the REGEN Dataset * The dataset enriches the existing Amazon Product Reviews archive by adding synthetic conversational elements, specifically targeting the gap in datasets that support natural language feedback. * **Critiques** are generated for similar item pairs within hierarchical categories, allowing users to guide the system by requesting specific changes, such as a different color or increased storage. * **Narratives** provide contextual depth through purchase reasons, product endorsements, and concise user summaries, helping the system justify its recommendations to the end-user. ### Unified Generative Modeling Approaches * The researchers framed a "jointly generative" task where models must process a purchase history and optional critique to output both a recommended item ID and a supporting narrative. * The **FLARE (Hybrid)** architecture uses a sequential recommender for item prediction based on collaborative filtering, which then feeds into a Gemma 2B LLM to generate the final text narrative. * The **LUMEN (Unified)** model functions as an end-to-end system where item IDs and text tokens are integrated into a single vocabulary, allowing one LLM to handle critiques, recommendations, and narratives simultaneously. ### Performance and Impact of User Feedback * Incorporating natural language critiques consistently improved recommendation metrics across different architectures, demonstrating that language-guided refinement is a powerful tool for accuracy. * In the Office domain, the FLARE hybrid model's Recall@10—a measure of how often the desired item appears in the top 10 results—increased from 0.124 to 0.1402 when critiques were included. * Results indicate that models trained on REGEN can achieve performance comparable to state-of-the-art specialized recommenders while maintaining high-quality natural language generation. The REGEN dataset and the accompanying LUMEN architecture provide a path forward for building more transparent and interactive AI assistants. For developers and researchers, utilizing these conversational benchmarks is essential for moving beyond "black box" recommendations toward systems that can explain their logic and adapt to specific user preferences in real time.

MUVERA: Making multi-vector retrieval as fast as single-vector search (opens in new tab)

MUVERA is a state-of-the-art retrieval algorithm that simplifies the computationally intensive process of multi-vector retrieval by converting it into a single-vector Maximum Inner Product Search (MIPS). By transforming complex multi-vector sets into Fixed Dimensional Encodings (FDEs), the system maintains the high accuracy of models like ColBERT while achieving the speed and scalability of traditional search infrastructures. This approach allows for efficient retrieval across massive datasets by leveraging highly optimized geometric search techniques that were previously incompatible with multi-vector similarity measures. ## The Limitations of Multi-Vector Retrieval While traditional models use a single embedding for an entire document, multi-vector models generate an embedding for every token, providing superior semantic depth but creating significant overhead. * Multi-vector representations lead to a massive increase in embedding volume, requiring more storage and processing power. * Similarity is typically calculated using "Chamfer matching," a non-linear operation that measures the maximum similarity between query tokens and document tokens. * Because Chamfer similarity is more complex than a standard dot-product, it cannot directly use sublinear search algorithms, often necessitating expensive exhaustive comparisons. ## Fixed Dimensional Encodings (FDEs) The core innovation of MUVERA is the reduction of multi-vector sets into a single, manageable vector representation that preserves mathematical relationships. * FDEs are single vectors designed so that their inner product closely approximates the original multi-vector Chamfer similarity. * The transformation process is "data-oblivious," meaning the mapping does not need to be trained on or adjusted for specific datasets or changes in data distribution. * By squeezing group information into a fixed-length format, MUVERA allows complex data points to be stored and queried using existing single-vector indexing structures. ## The MUVERA Retrieval Pipeline The algorithm functions as a multi-stage process that prioritizes both speed and precision through a retrieve-and-rerank architecture. * **FDE Generation:** Query and document multi-vector sets are mapped into FDEs to capture essential similarity information. * **MIPS-based Retrieval:** A standard MIPS solver indexes the document FDEs and rapidly identifies a set of likely candidates for a given query. * **Re-ranking:** The initial candidates are refined using the original, exact Chamfer similarity score to ensure the highest possible accuracy in the final results. MUVERA provides a practical framework for scaling high-accuracy multi-vector models to massive datasets without the traditional latency penalties. Its ability to bridge the gap between complex semantic modeling and optimized search infrastructure makes it a versatile tool for modern information retrieval systems.

Hosting the Tech Conference Tech-Verse (opens in new tab)

LY Corporation is hosting its global technology conference, Tech-Verse 2025, on June 30 and July 1 to showcase the engineering expertise of its international teams. The event features 127 sessions centered on core themes of AI and security, offering a deep dive into how the group's developers, designers, and product managers solve large-scale technical challenges. Interested participants can register for free on the official website to access the online live-streamed sessions, which include real-time interpretation in English, Korean, and Japanese. ### Conference Overview and Access * The event runs for two days, from 10:00 AM to 6:00 PM (KST), and is primarily delivered via online streaming. * Registration is open to the public at no cost through the Tech-Verse 2025 official website. * The conference brings together technical talent from across the LY Corporation Group, including LINE Plus, LINE Taiwan, and LINE Vietnam. ### Multi-Disciplinary Technical Tracks * The agenda is divided into 12 distinct categories to cover the full spectrum of software development and product lifecycle. * Day 1 focuses on foundational technologies: AI, Security, Server-side development, Private Cloud, Infrastructure, and Data Platforms. * Day 2 explores application and management layers: AI Use Cases, Frontend, Mobile Applications, Design, Product Management, and Engineering Management. ### Key Engineering Case Studies and Sessions * **AI and Data Automation:** Sessions explore the evolution of development processes using AI, the shift from "Vibe Coding" to professional AI-assisted engineering, and the use of Generative AI to automate data pipelines. * **Infrastructure and Scaling:** Presentations include how the "Central Dogma Control Plane" connects thousands of services within LY Corporation and methods for improving video playback quality for LINE Call. * **Framework Migration:** A featured case study details the strategic transition of the "Demae-can" service from React Native to Flutter. * **Product Insights:** Deep dives into user experience design and data-driven insights gathered from LINE Talk's global user base. Tech-Verse 2025 provides a valuable opportunity for developers to learn from real-world deployments of AI and large-scale infrastructure. Given the breadth of the 127 sessions and the availability of real-time translation, tech professionals should review the timetable in advance to prioritize tracks relevant to their specific engineering interests.

From research to climate resilience (opens in new tab)

Google Research is leveraging advanced artificial intelligence to transform climate science from theoretical exploration into scalable, real-world resilience tools. By developing sophisticated models for floods, cyclones, and hyper-local weather, the initiative provides critical lead times that empower communities to protect lives and livelihoods against increasingly frequent environmental threats. This transition from "impossible" research to global implementation highlights AI's capacity to bridge data gaps in the world's most vulnerable regions. ## AI-Powered Global Flood Forecasting * Google developed a global hydrological AI model, recently published in *Nature*, which enables riverine flood forecasts up to seven days in advance. * The system utilizes "virtual gauges" to analyze historical data and provide predictions in regions where physical water-monitoring infrastructure is non-existent. * The Flood Hub platform now covers over 100 countries and 700 million people, providing an expert data layer and API access for local governments and researchers. ## Cyclone Tracking and Intensity Prediction * Collaborative research between Google DeepMind and Google Research has produced models that predict storm existence, track, intensity, and size up to 15 days in advance. * The AI generates up to 50 different possible scenarios for each storm, providing a more nuanced view of potential impact than traditional physics-based supercomputer simulations. * Through the new Weather Lab website, these experimental models are being shared with the US National Hurricane Center to assist in forecasting during the Atlantic hurricane season. ## Global Nowcasting with MetNet-3 * The MetNet-3 state-of-the-art neural weather model provides hyper-local precipitation forecasts with a 5km resolution, updated every 15 minutes. * By utilizing satellite observations instead of traditional ground-based radar, the system delivers reliable weather data to regions like Africa that lack extensive physical infrastructure. * These 12-hour "nowcasting" windows are integrated directly into Google Search, specifically helping agricultural communities react to changing conditions to improve crop yields and reduce waste. These advancements demonstrate that the "art of the possible" is rapidly expanding, offering a future where data-scarce regions can access the same life-saving predictive capabilities as developed nations through global partnerships and satellite-based modeling.

A colorful quantum future (opens in new tab)

Google Quantum AI researchers have successfully implemented "color codes" for quantum error correction on the superconducting Willow chip, presenting a more efficient alternative to the standard surface code. This approach utilizes a unique triangular geometry to reduce the number of physical qubits required for a logical qubit while dramatically increasing the speed of logical operations. The results demonstrate that the system has crossed the performance threshold where increasing the code distance successfully suppresses logical error rates. ## Resource Efficiency through Triangular Geometry * Unlike the square-shaped surface code, the color code uses a hexagonal tiling arranged in a triangular patch to encode logical information. * This geometric configuration requires significantly fewer physical qubits to achieve the same "distance" (the number of physical errors needed to cause a logical error) compared to surface codes. * Experimental results comparing distance-3 and distance-5 color codes showed a 1.56× suppression in logical error rates at the higher distance, confirming the code's viability on current hardware. * While the color code requires more complex decoding algorithms and deeper physical circuits, recent advances in decoders like AlphaQubit have enabled the system to operate below the error correction threshold. ## Accelerating Logical Gates * Color codes allow for many single-qubit logical operations to be executed in a single step (transversal gates), whereas surface codes often require multiple error-correction cycles. * A logical Hadamard gate, for instance, can be executed in approximately 20ns using a color code, which is nearly 1,000 times faster than the same operation on a surface code. * Faster execution reduces the number of error-correction cycles an algorithm must endure, which indirectly lowers the physical qubit requirements for maintaining logical stability. * The research team verified these improvements through "logical randomized benchmarking," confirming high-fidelity execution of logical operations. ## Logical State Injection and Magic States * The researchers demonstrated a "state injection" technique, which is the process of preparing a physical qubit in a specific state and then expanding it into a protected logical state. * This process is essential for creating "magic states" (T-states), which are necessary for performing the arbitrary qubit rotations required for complex quantum algorithms. * By moving states from the physical to the logical level, the color code architecture provides a clear path toward executing the universal gate sets needed to outperform classical computers. While the color code currently exhibits a lower error suppression factor than the surface code, its advantages in hardware efficiency and gate speed suggest it may be the superior architecture for large-scale, fault-tolerant quantum computing as device hardware continues to improve.

Unlocking rich genetic insights through multimodal AI with M-REGLE (opens in new tab)

Google Research has introduced M-REGLE, a multimodal AI framework designed to analyze diverse health data streams simultaneously to uncover the genetic underpinnings of complex diseases. By jointly modeling complementary signals—such as electrocardiograms (ECG) and photoplethysmograms (PPG)—the method captures shared biological information and reduces noise more effectively than unimodal approaches. This integrated analysis significantly enhances the discovery of genetic associations and improves the prediction of cardiovascular conditions like atrial fibrillation. ## Technical Architecture and Workflow M-REGLE utilizes a multi-step process to transform raw physiological waveforms into actionable genetic insights: * **Multimodal Integration:** Instead of processing data types in isolation, the model combines multiple inputs, such as the 12 leads of an ECG or paired ECG and PPG data, to capture overlapping signals. * **Latent Representation Learning:** The system employs a convolutional variational autoencoder (CVAE) to compress these high-dimensional waveforms into a low-dimensional "signature" or latent factors. * **Statistical Refinement:** Principal component analysis (PCA) is applied to the CVAE-generated signatures to ensure the learned factors are independent and uncorrelated. * **Genetic Mapping:** These independent factors are analyzed via genome-wide association studies (GWAS) to identify significant correlations between physiological signatures and specific genetic variations. ## Improved Data Reconstruction and Genetic Sensitivity The transition from unimodal (U-REGLE) to multimodal modeling has led to substantial gains in both data accuracy and biological discovery: * **Error Reduction:** M-REGLE achieved a 72.5% reduction in reconstruction error for 12-lead ECGs compared to analyzing each lead separately, indicating a much higher fidelity in capturing essential waveform characteristics. * **Increased Discovery Power:** In a study involving over 40,000 participants from the UK Biobank, the multimodal approach identified 3,251 significant genetic loci associated with 12-lead ECGs, a notable increase over the 2,215 loci found by unimodal methods. * **Novel Findings:** The model identified specific genetic links, such as the *RBM20* locus, which were previously missed by standard clinical measurements but are known to be critical for heart muscle function. ## Interpretability and Disease Prediction Beyond identifying associations, M-REGLE offers generative capabilities that help clinicians understand the relationship between latent data and physical health: * **Waveform Synthesis:** By altering specific coordinates within the learned embeddings, researchers can observe how individual latent factors correspond to physical changes in a patient's ECG T-wave or PPG peaks. * **Clinical Utility:** The model identified specific embeddings (positions 4, 6, and 10) that distinguish patients with atrial fibrillation (AFib) from those without. * **Predictive Performance:** M-REGLE’s embeddings outperformed traditional clinical polygenic risk scores (PRS) in predicting AFib, demonstrating the value of incorporating raw waveform data into risk assessments. ## Practical Applications Researchers and clinicians can leverage M-REGLE to extract richer insights from existing biobank data and wearable device outputs. By integrating multiple modalities into a single analytical pipeline, the framework provides a more comprehensive view of organ system health, facilitating the identification of therapeutic targets and more accurate disease screening protocols.

Replacing the Database of a Payment System (opens in new tab)

The LINE Billing Platform team recently migrated its core payment database from Nbase-T to Vitess to address rising licensing costs while maintaining the high availability required for financial transactions. After a rigorous Proof of Concept (PoC) evaluating Apache ShardingSphere, TiDB, and Vitess, the team selected Vitess for its mature sharding capabilities and its ability to provide a stable, scalable environment on bare-metal infrastructure. This migration ensures the platform can handle large-scale traffic efficiently without the financial burden of proprietary license fees. ### Evaluation of Alternative Sharding Solutions Before settling on Vitess, the team analyzed other prominent distributed database technologies to determine their fit for a high-stakes payment system: * **Apache ShardingSphere:** While it offers flexible Proxy and JDBC layers, it was excluded because it requires significant manual effort for data resharding and rebalancing. The management overhead for implementing shard-key logic across various components (API, batch, admin) was deemed too high. * **TiDB:** This MySQL-compatible distributed database uses a decoupled architecture consisting of TiDB (SQL layer), PD (metadata management), and TiKV (row-based storage). Its primary advantage is automatic rebalancing and the lack of a required shard key, which significantly reduces DBA operational costs. * **Nbase-T:** The legacy system provided the highest performance efficiency per resource unit; however, the shift from a free to a paid licensing model necessitated the move to an open-source alternative. ### Vitess Architecture and Core Components Vitess was chosen for its proven track record at companies like YouTube and GitHub, offering a robust abstraction layer that makes a clustered database appear as a single instance to the application. The system relies on several specialized components: * **VTGate:** A proxy server that routes queries to the correct VTTablet, manages distributed transactions, and hides the physical topology of the database from the application. * **VTTablet:** A sidecar process running alongside each MySQL instance that manages query execution, data replication, and connection pooling. * **VTorc and Topology Server:** High availability is managed by VTorc (an automated failover tool), while metadata regarding shard locations and node status is synchronized via a topology server using ZooKeeper or etcd. ### PoC Performance and Environment Setup The team conducted performance testing by simulating real payment API scenarios (a mix of reads and writes) on standardized hardware (8vCPU, 16GB RAM). * **Comparison Metrics:** The tests focused on Transactions Per Second (TPS) and resource utilization as thread counts increased. * **Infrastructure Strategy:** Because payment systems cannot tolerate even brief failover delays, the team opted for a bare-metal deployment rather than a containerized one to ensure maximum stability and performance. * **Resource Efficiency:** While Nbase-T showed the best raw efficiency, Vitess demonstrated the necessary scalability and management features required to replace the legacy system effectively within the new cost constraints. ### Practical Recommendation For organizations managing critical core systems that require horizontal scaling without proprietary lock-in, Vitess is a highly recommended solution. While it requires a deep understanding of its various components (like VTGate and VTTablet) and careful configuration of its topology server, the trade-off is a mature, cloud-native-ready architecture that supports massive scale and automated failover on both bare-metal and cloud environments.