Loading…
Latest reads
Every post, summarised
Grounded summaries of engineering blogs, newest first. Each links to the original.
- 4,671 posts
- 20 companies
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Latest posts
Our Framework for Reviewing AI-Generated Code
JetBrains Research, in collaboration with Lund University, studied how to review AI-generated code as AI agents produce thousands of lines across multiple files quickly. The core problem identified is trust calibration: developers cannot gauge an LLM’s confidence per line, making line-by-line review unscalable. The study proposes a three-level conceptual framework: overview for high-level hypotheses, file-level risk stratification, and code snippet-level chunk decomposition with chain-of-thought linkage. This framework was developed through participatory design with 17 practitioners and a follow-up survey of 43 professionals, and will be presented at ESEIW 2026. The paper argues that current diff viewers are inadequate for AI-generated code and that tools must surface risk and confidence signals at the granularity where developers allocate attention.
Scaling and Operating a Large dbt Project on Databricks: IFCO's Data Team on Performance, Visibility, and Debugging
IFCO’s data platform team, with Databricks Forward Deployed Engineering, optimized a large dbt project on Databricks to handle billions of daily tracking events for reusable packaging assets. The pipeline faces scale, unpredictable late-arriving data, and the need to reconcile history when late events change previously reported asset timelines. The team tuned incremental model settings to match Delta Lake write behaviors, including liquid clustering, deliberate merge vs. delete+insert strategies, and bounding writes to recent ingestion windows. Diagnosing performance from actual query plans rather than compiled SQL revealed that the consolidation model was scanning far more rows than necessary. These changes cut the core semantic layer job’s daily runtime by over 60 percent, retired a costly nightly full refresh, and reduced daily compute cost by 58 percent.
Java and Kotlin by IntelliJ IDEA: The LSP-Based Extension Reaches Release Candidate
JetBrains announced that Java and Kotlin by IntelliJ IDEA, an LSP-based extension, has reached Release Candidate stage. The extension brings IntelliJ IDEA’s Java and Kotlin language intelligence to third-party editors like VS Code and Cursor. Since the preview, it has been downloaded over 20,000 times and includes features such as smart code completion, navigation, inspections, quick-fixes, project-wide refactorings, DAP-based debugging, Maven, Gradle, and Bazel project support, built-in test running, and Spring tooling interoperability. The RC is the final EAP build and will be available for 30 days without a license; after the stable release, a 30-day free trial will be offered, then an IntelliJ IDEA Ultimate subscription is required. The extension also makes its intelligence available to coding agents in Cursor, enabling semantic code search and analysis.
Content and Community Basics for JetBrains Content Creators
The post covers community-building basics for JetBrains Content Creators in a golden age of digital creation with high demand and AI adoption, accepting new creators daily. Creators should include the JetBrains promotional discount code in video descriptions, with an example offering 20% off for new and existing subscriptions, and they can explore all JetBrains IDEs through the All Products Pack to diversify their workflow. Maintaining active social media is essential for distribution, and the program supplies news-worthy items like merch drops and IDE updates for creators to share with their audience. Staying in touch through communication, challenges, and giveaways is crucial, and the program provides up to twelve free IDE subscriptions per creator for giveaways. Even a simple subscription plus repost formula works effectively, and all new creators are welcome to join the program.
JetBrains Joins the Open Source Security Foundation
JetBrains has joined the Open Source Security Foundation (OpenSSF) as a general member. OpenSSF is a cross-industry Linux Foundation initiative that works to secure the software supply chain, hosting projects like Sigstore for signing and SLSA for supply chain integrity, and its AI/ML Security Working Group publishes guidance on signing ML models and safe instructions for AI code assistants. The foundation announced the membership at OpenSSF Community Day Europe in Prague. JetBrains cites the rise of coding agents and AI as creating new security challenges, making it more important for developers to understand and verify code and dependencies, and notes that its own product JetBrains Air addresses these questions. The company aims to collaborate with the industry through OpenSSF to help shape the future of secure software development, as stated by Katherine Druckman, Head of Community and Partnership Engagement.
Unreal MCP now speaks Sentry
Unreal Engine 5.8 introduces Unreal MCP, an experimental feature that runs a first-party MCP server inside the editor, allowing coding agents to drive the editor directly. The Sentry plugin registers tools and skills that agents can use to set up crash reporting, including DSN configuration, symbol upload, instrumentation, and crash debugging. Agents can discover and use these tools through the editor’s MCP interface, with skills that follow ground rules like asking before changing settings and never exposing secrets. The integration also supports connecting to Sentry MCP for server-side operations like fetching DSNs and looking up events. In a test with the Sentry Tower demo, an agent configured Sentry from scratch, enabled symbol upload, packaged a build, captured a crash, and identified the root cause as an empty projectile class in a level-up button.
Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance
Falcon-Emirati-7B is a dialect-specialized language model built on Falcon-H1-Arabic to understand and generate Emirati Arabic, a Gulf dialect with distinct vocabulary and cultural context that Modern Standard Arabic models miss. The model uses a hybrid architecture combining Mamba state-space models and Transformer attention running in parallel inside every block, with fused outputs for efficient long-context processing. Training required a dedicated data pipeline of authentic Emirati web content, MSA material about Emirati culture, and synthetic data guided by strict glossaries and style rules to cover gaps in available dialectal text. On the Alyah benchmark, Falcon-Emirati-7B scored 84.83%, outperforming larger multilingual models, and achieved 85.57% on the UAE cultural understanding task in ArabCulture-Dialogue. The 7B parameter size was chosen as the best balance of quality against training and inference cost for dialect adaptation.
Smarter personalization: How property data helps us understand user price sensitivity
Existing systems estimate user price sensitivity from spending behavior or demographic proxies, missing residential property values and creating one-dimensional profiles. This invention proposes a four-step method that combines public property data with LLM-assisted data structuring, building-level price estimation, hierarchical clustering, and user mapping. The method collects public property records, uses LLMs to standardize attributes such as postcode, transaction price, and property size, then aggregates transactions within buildings to calculate weighted average sale prices. A hierarchical clustering model groups buildings into affluence tiers based on estimated resale values, accounting for regional market differences, and users are mapped to these tiers via aggregated geographic signals. The approach remains unimplemented and untested in live markets, with potential applications in personalization, advertising, and operational planning subject to future technical validation and privacy approvals.
Data Modeling and Schema Design for MongoDB Search Part 1: Practical Considerations
Data modeling for MongoDB Search requires balancing conceptual, logical, and physical schema decisions against workload demands and system constraints. The $search stage must be the first in an aggregation pipeline, queries a single Search index, and runs against a single collection, with results returned in relevancy order. The underlying Lucene engine supports two mappings for arrays of sub-documents: default document mapping flattens fields into one index object, while embeddedDocuments mapping creates separate Lucene index objects per sub-document. Embedded mapping enables cross-field matching within sub-documents but explodes the index object count and causes heavy reindexing on any update, as the entire block is deleted and re-added. This post establishes the foundational concepts and physical constraints that will inform schema patterns in Part 2.
Data Modeling and Schema Design for MongoDB Search Part 2: Use Cases and Patterns
The post presents best practices for data modeling and schema design in MongoDB Search, covering use cases and patterns such as the Entity Pattern to guide developers in structuring collections for effective search. It emphasizes that searchable items, like movies, cast members, or genres, should be modeled as first-class documents in their own collections rather than as nested structures or ambiguous attributes. The Entity Pattern addresses ‘find anything anywhere’ and as-you-type suggestions by representing all domain entities as documents in a specialized collection with stable identifiers and a type field for filtering. The Attribute Pattern is explicitly discouraged for search because its key-value structure prevents type-sensitive field mapping, making it better to model attributes as straightforward fields. The post details a multi-indexing strategy on the name field that applies left-edge, phonetic, shingled, and language-agnostic analysis to enable flexible matching and relevancy scoring, with optional type-specific metadata nested in sub-documents.
My Voice AI Demo Was Hit by a $982 Traffic-Pumping Attack! Here’s What I Learned
A voice AI demo using Twilio Conversation Relay and FastAPI was exploited when automated clients repeatedly called its unauthenticated /api/call endpoint, generating 1,008 requests and 921 outbound calls to Israeli +972229 numbers. The incident cost $971.50 in Twilio voice usage and $10.58 in Google Gemini fees, matching International Revenue Sharing Fraud that keeps calls connected for per-minute revenue. The author used the Twilio CLI to group call records showing 920 of 921 calls targeted the same narrow Israeli range and correlated Azure Log Analytics to confirm the exploited endpoint. Calls were kept alive by automated scripts replaying conversation-extension prompts from only 541 distinct lines across 8,363 transcriptions, with some recordings mentioning happy trafficking. The post recommends authentication, duration limits, geographic permissions, signature validation, usage triggers, and demo isolation as essential controls for public voice AI apps.
How to design an AI agent for resolution instead of rapport
Most conversation design guidance focuses on voice and tone, but consumers prioritize immediate issue resolution (40%), speed (40%), accuracy (38%), and 24/7 availability (38%) over rapport. The post recommends opening with capability, loading context before the first turn using persistent profiles, and designing a second option when the first fails. Escalation should be triggered by conversation rules like a second failed attempt or explicit request, not treated as an error condition. Confirmation turns should be cut except for irreversible actions, and agents should clearly state what they cannot do. An agent with customer history, a backup path, and a clean exit is more helpful than one with a carefully written persona but lacking those features.
RISED: Rubrics for Agentic Multi-Environment Selection and Self-Distillation
Training a single LLM agent jointly across diverse interactive environments is challenged by scalar rewards that provide limited information about cross-environment relationships and no within-group reward contrast when rewards are identical. RISED addresses this by using rubrics—textual descriptions of rollout behaviors—to guide both online data selection and policy supervision. An LLM judge tags each rollout with a predefined rubric vocabulary shared across environments, and these profiles select data aligned with the batch’s behavioral composition while limiting overlap with already-selected data. Positive rubrics supply token-level supervision via an on-policy self-distillation teacher, while negative rubrics steer subsequent rollouts away from recurring failure modes. Across model backbones, RISED achieves the highest mean pass rate across environments and ranks first or second in every individual environment.
Remote control for local agents
Cursor has introduced Remote Control, allowing users to view and interact with local agents running on their computer from the Cursor iOS app; agents appear in a list and users can tap one to see its activity or send a message. The feature is enabled by default for all users except Enterprise organizations; users tap their computer in the app and approve a pairing request in the Cursor desktop app. Agents remain running on the user’s computer, which must stay on and online; the app provides a “Keep this computer awake” setting to prevent sleep, and the computer must be plugged in with the lid open. Enterprise admins can enable Remote Control in Org settings under Security and identity. Remote Control does not rely on cloud agents and is available now in the Cursor iOS app.
Expanding our enterprise inference capacity with IBM Cloud and NVIDIA
Together AI is collaborating with IBM and NVIDIA to scale enterprise-grade AI inference using a dedicated cluster of NVIDIA B300 GPUs on IBM Cloud with NVIDIA Spectrum-X Ethernet networking. This is the first large-scale inference cluster of its kind on IBM Cloud, and Together AI is the first customer running on it. Together AI operates the inference layer, IBM provides the cloud, and NVIDIA delivers the silicon and networking. The company serves hundreds of trillions of tokens per month to over a million developers, and demand continues to climb. Enterprises choose open models to keep their data sovereign and achieve frontier-level performance at a fraction of closed-model cost.
Data-Driven, Interoperable Core Banking: A BIAN and MongoDB Perspective
Core banking modernization is hindered by semantic fragmentation, where the same banking concepts are defined differently across systems, creating duplicated logic and fragile integrations. The Banking Industry Architecture Network (BIAN) provides a shared semantic foundation by defining common banking domains, boundaries, and service interactions, enabling a composable architecture. MongoDB’s document model, using JSON as both storage and interchange format, allows banks to represent business domains flexibly and evolve incrementally without schema migrations, eliminating the ORM impedance mismatch. MongoDB ensures financial integrity through single-document atomicity for common ledger updates and multi-document ACID transactions for complex cross-domain workflows, while Change Streams enable event-driven operations. Together, BIAN and MongoDB create an operating model for continuous transformation, where teams modernize domain by domain, supported by a shared business language and a scalable, flexible data platform.
EmbeddingGemma 2: The Developer Guide
Modern search and RAG applications need to work across diverse content types, but finding models with strong retrieval accuracy and low latency without massive compute is challenging. EmbeddingGemma 2 is a compact open model under Apache 2.0, based on Gemma 4, that maps text, code, images, video, and audio into a unified 768-dimensional space with a modular architecture scaling from 270M to 740M parameters. Key capabilities include native multimodal retrieval, superior code understanding, modular memory footprint, and flexible vector storage via Matryoshka Representation Learning enabling dynamic truncation from 768 to 128 dimensions. Under the hood, it uses modular encoders—a text/code decoder, vision encoder, and audio encoder—that project into a shared space and can be loaded selectively to minimize memory. The developer guide demonstrates how to use sentence-transformers to load the model, embed with task prompts, handle media inputs, truncate dimensions, and choose configurations based on storage and recall needs.
Bring multimodal semantic search to the edge with EmbeddingGemma 2
Google DeepMind launched EmbeddingGemma 2, an open-weight multimodal embedding model that maps text, images, video frames, and audio into a single vector space. Designed for local, privacy-first applications, it runs on as little as ~191MB active RAM for text-only weights and ~567MB for the full multimodal model on a Google Pixel 11 Pro. The model powers two new interactive showcases in Google AI Edge Gallery: Instant Media Search, which retrieves local images and videos via natural language or example images, and Video Moments Finder, which locates specific visual moments in local video files without transcribing audio. Google AI Edge Foresight for Mac uses EmbeddingGemma 2 alongside Gemma 4 for local meeting note-taking, cross-modal retrieval, and live question answering. Optimizations include modular encoders, quantization-aware training to INT4/INT8, and Matryoshka Representation Learning slicing to reduce storage footprints by up to 8x.
Turning Customer Identity into A Growth Engine
Customer Identity and Access Management (CIAM) has traditionally been viewed as a security obligation, but Auth0 argues it can be a growth driver when optimized for conversion. The Auth0 Identity Conversion Suite combines Anonymous Sessions and Experiment Center to transform identity from a rigid gatekeeper into a dynamic growth engine. Anonymous Sessions issue scoped access tokens to track anonymous users from their first interaction, preserving cart contents, preferences, and consent selections when they convert to authenticated accounts. Experiment Center enables native A/B testing and gradual rollouts of authentication flows, allowing teams to test changes on controlled traffic segments before global deployment with audit trails. The suite addresses high cart abandonment in retail, cross-device context loss in travel, and onboarding friction in financial services by enabling seamless guest transitions, cross-channel journey unification, and progressive trust and compliance.
NEAREST BY Join: Scaling Vector Search in Databricks Runtime
Batch vector search workloads, such as entity resolution and semantic tagging, require high throughput across millions of queries against billions of vectors, unlike real-time serving. Databricks built NEAREST BY, a first-class top-k ranking join syntax, directly into its runtime engine rather than federating to an external endpoint. The feature is implemented as a logical join that the optimizer lowers onto standard relational operators, enabling fault tolerance and elastic parallelism. A fused Photon operator collapses the plan’s middle section with a blocked GEMM kernel designed to approach peak arithmetic intensity for vector scoring. An optional IVF index, stored as an ordinary liquid-clustered Delta table, allows approximate queries to probe only relevant clusters while preserving the same primitives and correctness guarantees.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.