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
What's new in the ClickHouse .NET Driver: the road from 1.0 to 1.3
The ClickHouse .NET Driver moved from its 1.0 stable release to versions 1.1–1.3, changing how applications insert, read, and parameterize data while expanding type coverage, configurability, observability, and .NET ecosystem integrations. The largest ergonomic change is registered POCO support: RegisterPocoType() validates mappings up front, pre-compiles getters and setters, and enables InsertBinaryAsync with IEnumerable plus lazy QueryAsync() results through IAsyncEnumerable. Attributes customize column and ClickHouse type mappings, JSON serialization supports POCOs with paths and ignored properties when JsonWriteMode is Binary, and three optional extension points govern parameter type inference, wire formatting, and read conversion. The announced 1.4 roadmap targets allocation reductions, pluggable compression, and box-free POCO operations; version 1.5 is planned to add a Native/TCP client with columnar transfer and server progress information.
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Sentence Transformers v6.0 adds MultiVectorEncoder, a fourth model type for ColBERT-style late-interaction retrieval, alongside a training approach for finetuning existing models or building them from base transformers. Multi-vector models retain one small vector per token and score query-document pairs with MaxSim, where each query token selects its best document-token match and the scores are summed. The guide covers models, datasets, loss functions, training arguments, evaluators, trainers, callbacks, and multi-dataset training, with examples using pip install -U "sentence-transformers[train]". On a medical retrieval evaluation, the finetuned multi-vector-encoder/mLateOn-medical model, trained for 14.5 hours on a single RTX 3090, outperformed every general-purpose retrieval model tested across dense, sparse, lexical, and multi-vector systems.
Building a data plane from scratch: Stripe’s own high-performance distributed proxy
Stripe replaced Envoy in its service mesh after reliability, connection-cardinality, rollout-risk, and customization problems became increasingly difficult at its scale. Mesh-proxy, a Go proxy, implements HTTP/1.1 and HTTP/2, connects directly to Stripe’s service registry, and supports retries, hedging, rate limiting, and dynamic priority-based routing. Unlike Envoy’s worker-per-core model, it uses one read and one write goroutine per connection and multiplexes requests over a shared connection, reducing connection counts from one per worker to one per host. Load tests and deployment results showed about half the CPU under high load and roughly 50% lower latency, equivalent to around 3,000 days of cumulative inter-service latency saved daily. Direct registry integration also reduced unnecessary xDS updates, while custom Ruby balancing added 15–20% headroom before bursts overwhelm machines; Stripe is now pursuing anomaly detection and smart routing.
Migrating to Auth0 for Startups from Another Auth Solution
Evenpay develops compensation transparency software to meet the EU Pay Transparency Directive, requiring strict safeguards for sensitive employee salary data. The startup initially selected a lightweight B2B authentication platform for quick onboarding, but soon encountered unmet data residency and compliance demands from Finnish enterprise clients. To resolve these limitations, Evenpay migrated all customers to Auth0 over a few weeks using documentation and AI assistance. Adopting Auth0 allowed the team to fulfill regional EU data controls while reducing managed Single Sign-On setup times from weeks to as little as one day through self-service options. Furthermore, relying on an established authentication brand provided instant credibility with enterprise security officers without requiring custom authentication development.
PROOF-Gen: From Optimized Data to Better Distillation
Standard distillation pipelines for tool-calling agents use generate-and-filter supervised fine-tuning on teacher trajectories, discarding failed trials and missing training signals on hard scenarios. On τ 2-bench, 57% of teacher trials fail, with two-thirds being near-misses undone by a single error. PROOF-Gen (Per-scenario Reflective Optimization to Overcome Failed Generation) addresses this loss by employing a reflector that reviews execution traces and evaluation feedback to craft corrective guidance for the teacher. Once passing trajectories are generated, this scaffold guidance is stripped so students train on clean demonstrations. Per-scenario prompt optimization recovers 93% of failed scenarios on τ 2-bench, boosting Qwen3-4B-Instruct-2507 Passˆ1 from 0.132 to 0.529 and lifting deployed on-device model performance across all locales.
IDEA Prune: An Integrated Enlarge-and-Prune Pipeline in Generative Language Model Pretraining
Recent advancements in large language models necessitate efficient deployment options within constrained inference budgets. While structured pruning offers better token efficiency than training target-sized models from scratch, previous work frequently ignores the pretraining of enlarged models. To resolve this gap, researchers developed an integrated enlarge-and-prune pipeline that unifies enlarged model training, pruning, and recovery under a single cosine annealing learning rate schedule. This system incorporates an iterative structured pruning technique to gradually eliminate parameters, redistributing model capacity across surviving neurons and avoiding the knowledge loss seen in naive approaches. Evaluated on compressing 2.8B parameter models to 1.3B across up to 2T pretraining tokens, the approach yields superior downstream model performance.
Luce: Relightable Gaussians for 3D Asset Generation
High-fidelity image-to-3D asset generation requires 3D representations that capture geometry alongside physically based rendering modalities for seamless pipeline integration and relighting. Luce addresses this challenge by unifying geometry and material properties into a voxelized multimodal Gaussian cloud with dedicated primitives for albedo, metallic-roughness, and surface normals. A variational autoencoder compresses these multimodal primitives into a unified material-aware latent space. A rectified-flow transformer subsequently generates this latent from a single input image using multi-layer features extracted from a pretrained image encoder. On the Toys4K dataset, Luce improves FID by 28% over the leading baseline while maintaining fine details such as text and logos.
Restore LLM Inference Capacity in Seconds with Shadow Engine Recovery in NVIDIA Dynamo
LLM engine process failures typically trigger cold restarts that reload weights into HBM, compile kernels, and recapture CUDA graphs, leaving surviving workers to absorb displaced traffic for minutes. Shadow engine recovery, a preview feature in NVIDIA Dynamo, keeps a fully initialized standby on the same GPUs as the active engine. GPU Memory Service (GMS) preserves one physical weight copy across engine processes, while the parked shadow precomputes its CUDA context, graphs, communicators, and mappings but defers KV cache materialization. In a GLM-5.2 test, the shadow resumed service in 7.3 seconds after a worker failure, compared with 283 seconds for a cold restart, reducing post-fault p50 TTFT from 23,815 ms to 1,311 ms and improving decode rate from 12 to 46 tok/s/user. The preview does not cover hardware, node, or multi-node failures and currently starts promoted shadows with empty KV caches.
How Capital Markets Finance Protects Balance Sheet Returns
Capital markets finance must protect returns on a balance sheet exposed continuously to market, funding, and liquidity risk, while funding costs, capital consumption, and liquidity needs can change intraday. Fragmented desk P&L, treasury, and risk data require shared business context, with ontology and governance linking numbers to desks, products, positions, funding, capital, and permissions. Databricks’ Genie One is presented as a data-smart AI coworker that lets authorized finance users investigate governed data, trace emerging funding or profitability issues, and surface the definitions and source evidence behind capital and valuation figures. One top global bank reduced regulatory data processing for liquidity reporting from 10 hours to 8 minutes, while pricing, funding, risk, and capital decisions remain with accountable people.
How Suprema Gaming made its data platform agent-ready with ClickHouse Cloud
Suprema Gaming migrated its analytics platform from Snowflake to ClickHouse Cloud on AWS as part of a company-wide move toward agentic operations. The previous stack delivered data about four hours late, required heavy dbt rebuilds, and often took minutes to answer queries, limiting real-time investigations and interactive analysis. Suprema rebuilt ingestion with ClickPipes for supported sources and Debezium plus Kafka for others, while dbt and GitHub Actions deploy the warehouse-layer structure without external cron jobs, Airflow, or a separate orchestrator; governed access uses role-based controls and query-time PII masking. Before cutover, every metric was validated against the reference source, and the resulting platform reduced warehouse spend by roughly 62%, improved direct-read performance 27x, and maintains about one-minute p95 freshness, supporting an early production deployment of ClickHouse Agents.
Project Lighthouse — Part 3: Introducing project-lighthouse-anonymize
Airbnb has open-sourced project-lighthouse-anonymize, a Python library powering their privacy-preserving anonymization used to measure user experience disparities. The release is accompanied by technical papers detailing Core Mondrian, an extended partition-based anonymization algorithm, alongside a framework for measuring anonymized data quality. The library enforces k-anonymity and p-sensitive k-anonymity while offering built-in data quality metrics.
Funding better evaluations of AI’s impact on wellbeing
A $5 million grant program will fund independent research into how AI affects user wellbeing, offering direct funding, model access, and technical support for open-source evaluations. The initiative addresses the difficulty of judging wellbeing-related behavior, where risks may emerge only across long, shifting conversations and where otherwise reasonable advice can become harmful in a different context, such as disordered eating. Grantees will work independently and publish reusable evaluations or benchmarks, with guidance asking projects to define pass/fail criteria, involve clinical and subject-matter experts, test overcompliance and overrefusal, model realistic multi-turn scenarios, and validate graders against experts. Applications are due September 21, with selected full-proposal applicants notified by October 5.
Choosing Data Governance Tools for Enterprise Data Governance
Enterprise data governance tools address fragmented data estates in which teams struggle to locate assets, identify ownership, protect sensitive information, and maintain consistent quality across warehouses, lakes, SaaS applications, and spreadsheets. Positioned above storage and compute, these tools read metadata, schemas, and query logs to provide a shared governance layer for cataloging, lineage, classification, policy enforcement, quality monitoring, and audit reporting. The guide distinguishes standalone catalogs, point solutions, enterprise suites, platform-native tools, and open-source options while treating AI and agent governance as an increasingly important capability for models, prompts, and autonomous agents. It recommends evaluating candidates against seven criteria, comparing total cost of ownership, and planning a phased rollout that starts with highest-risk domains; a lakehouse-native approach can reduce synchronization between separate governance systems.
Introducing the Anthropic Economic Index
Anthropic introduced the Anthropic Economic Index to track how AI is used in work and to study its effects on labor markets over time. Its first report analyzes approximately one million anonymized Free and Pro Claude.ai conversations, using Clio to match them with occupational tasks from the U.S. Department of Labor’s O*NET database. Usage concentrated in computer and mathematical work, which represented 37.2% of queries, while 36% of occupations showed AI use in at least a quarter of associated tasks and about 4% reached three-quarters. The report found 57% augmentation versus 43% automation, lower use in both low- and very-high-wage occupations, and plans for open data and repeated analyses, while noting possible classification errors and coding overrepresentation.
Launching the Anthropic Economic Futures Programme in the UK and Europe
Anthropic is expanding its Economic Futures Programme to the UK and Europe to support research and policy discussions about AI’s effects on labor, productivity, and growth. The expansion begins with an Economic Futures Symposium at the London School of Economics and Political Science, where more than 100 experts and selected researchers will discuss policy proposals. Planned support includes research grants, API credits for UK and EU researchers, forums for evidence-based policy, and more granular data on European AI use. The Anthropic Economic Index reports that coding is Claude’s most common use across Europe, while UK usage most often supports academic research, writing, and educational content; Germany shows high per-capita use for manufacturing equipment, and France records four times the global average for dining, shopping, and nightlife information. Regular public releases will track adoption across European industries and regions as the programme seeks to inform decisions about the workforce transition.
Supporting ambitious external research through the Anthropic Economic Futures Research Fund
The Anthropic Economic Futures Research Fund commits $200 million to external research on interventions that could help society prepare for AI’s economic impacts. It prioritizes five areas: workplace integration, support for AI-driven transitions, income support for displacement, worker stakes in AI-driven growth, and evidence on public investments. The fund will favor large-scale randomized controlled trials, ambitious pilots, and program evaluations, with particular interest in projects that share findings at key milestones and can scale rapidly if promising; typical awards will be $5 million to $30 million, with nothing below $1 million funded directly. Eligible applicants include accredited degree-granting institutions, independent research and policy organizations, and nonprofits with experience running field experiments at scale, while the fund remains global despite US-centric examples.
Anthropic Economic Index: Insights from Claude 3.7 Sonnet
Anthropic’s second Economic Index report examines how Claude.ai usage changed after Claude 3.7 Sonnet launched, using 1 million anonymized Free and Pro conversations from the following 11 days. Clio maps conversations to 17,000 tasks in the U.S. Department of Labor’s O*NET database, enabling comparisons across occupations and interaction modes, while the report also introduces a 630-category bottom-up taxonomy. Coding, education, and science usage increased modestly, and extended thinking appeared most often in technical and creative work, including computer science research, software development, multimedia art, and video game design. Augmentation remained 57% of usage, learning interactions rose from about 23% to 28%, and the report releases task- and occupation-level datasets covering thinking mode and augmentation-versus-automation patterns.
The Anthropic Economic Index connector
The Anthropic Economic Index connector for Claude makes the Index’s data available for direct question-and-answer exploration. The Index measures how AI is used in the economy, including occupations, geographic usage, and the tasks people perform or automate with Claude. Users enable the connector from the claude.ai connectors directory, without installing anything, then can ask broad or specific questions and request the underlying data behind an answer. Responses are grounded in Index data, but the source cautions that it reflects Claude usage patterns rather than the labor market as a whole; the full datasets remain freely available on the website for further exploration.
Data Mesh vs. Data Fabric: Key Differences and How the Lakehouse Resolves the Debate
Data mesh and data fabric address different constraints in distributed data architecture: mesh changes ownership and accountability, while fabric automates integration and governance. A data fabric uses active metadata, machine learning, centralized policy enforcement, and virtualized access to discover, govern, and connect data across heterogeneous environments with less data movement. A data mesh assigns domain teams responsibility for data products, quality, reliability, and lifecycle management, supported by self-serve infrastructure and federated computational governance. The article presents a lakehouse as a substrate where domains publish products while capabilities such as Unity Catalog, Delta Sharing, and Lakehouse Federation provide metadata, governance, and sharing. Its conclusion is that the approaches are complementary: organizations should identify whether organizational bottlenecks or technical fragmentation is the dominant constraint, then consider a hybrid.
Granite 4.2 LLMs: How They're Built
Granite 4.2 is IBM’s first family of dense, decoder-only reasoning language models, offered in 3B, 8B, and 30B sizes. The models use a shared design built around Grouped Query Attention, Rotary Position Embedding, SwiGLU, RMSNorm, bfloat16 precision, and separate input/output embeddings; each has a 131,072-token sequence length. Pre-training starts from scratch on approximately 15 trillion tokens across five phases, with the final phase extending the context window to 512K tokens. Supervised fine-tuning uses about 7.2 million samples spanning agentic and non-agentic data, followed by multi-stage reinforcement learning; the 8B and 30B models also receive agentic RL in sandboxed environments for tool use, code execution, terminal control, and web search. All models support thinking, non-thinking, and low-effort thinking modes, native tool calling, and Apache 2.0 licensing.
Top topics
- Architecture834
- Performance816
- LLMs660
- Machine Learning622
- Open Source563
- AI Agents553
- AI470
- Developer Experience448
Weekly digest
The best engineering reads, once a week.