--- title: "Latest reads" description: "The engineering internet, summarised so you can actually read it." --- # Latest reads > The engineering internet, summarised so you can actually read it. ## Articles ### [Building a Hyper Self-Service, Distributed Tracing and Feedback System for Rule & Machine Learning (ML) Predictions](https://yomu.fyi/post/building-a-hyper-self-service-distributed-tracing-and-feedback-system.md) - Company: [Grab](https://yomu.fyi/company/grab.md) - Author: Warren Zhou - Published: May 24, 2021 Grab's Trust, Identity, Safety, and Security team processes billions of daily rule and machine learning decisions for fraud detection, safety, and identity checks. Earlier logging approaches using plain text Kibana logs and the ActionTrace library lacked structured formats, dynamic entity customization, and fine-grained access controls. To resolve these limitations, the team built Archivist, a centralized tracing, statistics, and feedback system. Archivist ingests events through an SDK into Kafka streams, buffers and routes data into Elasticsearch indices and Amazon S3, and provides a role-based user portal. The platform handles 80 million daily logs across roughly 50 business scenarios, reducing scenario onboarding times from days to minutes. ### [Our Journey to Continuous Delivery at Grab (Part 2)](https://yomu.fyi/post/our-journey-to-continuous-delivery-at-grab-part-2.md) - Company: [Grab](https://yomu.fyi/company/grab.md) - Author: Sylvain Bougerel - Published: May 10, 2021 Conveyor, an in-house continuous delivery system at Grab, introduces hermetic deployments by tracking application code alongside static and dynamic configuration parameters. This hermeticity guarantees that production releases use combinations of versioned artifacts previously verified in staging, preventing rollback incompatibilities. Conveyor replaces single multi-environment pipelines with decoupled pipelines, while automating cluster locking, deployment slot scheduling, release note generation, and canary monitoring with automated rollbacks. These automations reduced production deployment failure rates from 1.5% to an average of 0.3% over a three-month period. Grab also doubled the volume of production changes between 2018 and 2020 while saving more than 5,000 man-days of engineering effort in 2020. ### [Supabase Beta April 2021](https://yomu.fyi/post/supabase-beta-april-2021.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: May 5, 2021 Supabase focused on platform stability, security enhancements, and community support during its April 2021 beta update cycle. To address visual accessibility needs for developers with astigmatisms or those operating in brightly lit environments, the team delivered Light Mode within the dashboard settings. The release also introduced OpenAPI-powered Storage API documentation, community translations for the main repository spanning nineteen distinct languages, and an experimental synchronization server connecting Postgres with Stripe. Dashboard components are systematically migrating into an open-source UI library, establishing necessary foundations for local development tooling and self-hosted environments. Additionally, the team reported substantial contributor growth across its realtime server and UI repositories while continuing engineering on a Workflows engine aimed at a July release. ### [Scaling-up BERT Inference on CPU (Part 1)](https://yomu.fyi/post/scaling-up-bert-inference-on-cpu-part-1.md) - Company: huggingface.co - Author: Morgan Funtowicz - Published: Apr 20, 2021 Scaling BERT-like model inference on CPUs presents practical deployment challenges as transformer architectures proliferate in production environments. Using a reproducible Hydra-based benchmarking framework, inference workloads were evaluated on an Amazon Web Services c5.metal instance featuring an Intel Xeon Platinum 8275 processor with AVX512 and VNNI support. Out-of-the-box evaluations showed PyTorch 1.8.1 delivering lower latency than Google TensorFlow 2.4.0, likely due to PyTorch relying on OpenMP and Intel MKL/oneDNN rather than Eigen. Implementing multiple inference streams by pinning parallel model instances to dedicated CPU cores yielded near-linear throughput scaling. Matching instance core allocations to specific problem sizes also demonstrated that smaller workloads can achieve optimal latency on fewer cores, enabling up to a sixfold reduction in cloud compute costs. ### [How We Improved Agent Chat Efficiency with Machine Learning](https://yomu.fyi/post/how-we-improved-agent-chat-efficiency-with-machine-learning.md) - Company: [Grab](https://yomu.fyi/company/grab.md) - Author: Suman Anand - Published: Apr 19, 2021 Agent typing time represented a large portion of Grab's chat support journey, and 85% of messages were still free typed because agents customized static templates to fit their personal style. To accelerate typing across multilingual markets without robotic templates, Grab built SmartChat, a machine learning feature that provides contextual sentence completion. The team opted for a lightweight seq2seq architecture using single-layered GRU encoder-decoders in TensorFlow instead of bulky attention models to keep model latency under 100ms. The user interface was implemented in React using a content-editable div with inline typeahead suggestions activated via keyboard shortcuts. ### [Introducing 🤗 Accelerate](https://yomu.fyi/post/introducing-accelerate.md) - Company: huggingface.co - Author: Sylvain Gugger - Published: Apr 16, 2021 🤗 Accelerate enables PyTorch developers to execute raw training loops across CPUs, multi-GPU setups, and TPUs with minimal boilerplate modifications. Standard distributed training typically demands manual device assignments, DistributedSampler configurations, and divergent code paths that break single-device portability. By wrapping models, optimizers, and dataloaders through a unified prepare call, Accelerate automatically manages hardware placement, mixed precision operations, and sampler batch slicing without requiring custom distributed samplers. Distributed evaluation is streamlined using a gather utility that consolidates prediction tensors across active worker processes. Furthermore, a dedicated command-line interface provides an interactive questionnaire to store runtime configuration defaults and orchestrate local or AWS SageMaker runs. ### [Distributed Training: Train BART/T5 for Summarization using 🤗 Transformers and Amazon SageMaker](https://yomu.fyi/post/distributed-training-train-bart-t5-for-summarization-using-transformer.md) - Company: huggingface.co - Author: Philipp Schmid - Published: Apr 8, 2021 Hugging Face and Amazon SageMaker introduced optimized Deep Learning Containers alongside a SageMaker Python SDK estimator to streamline distributed transformer model training. This workflow demonstrates fine-tuning the 400-million-parameter facebook/bart-large-cnn model for dialogue summarization on the 16,000-conversation samsum dataset. Using SageMaker Data Parallelism enabled in the distribution configuration, training executes across two ml.p3dn.24xlarge compute instances equipped with sixteen total GPUs and a cumulative batch size of 64. The HuggingFace estimator pulls the run\_summarization.py script directly from GitHub, enables mixed precision with fp16, and tracks evaluation performance through ROUGE scores. Following completion, the unzipped model artifacts and a generated model card are pushed to the Hugging Face Hub using the huggingface\_hub library for hosted inference. ### [Supabase Beta March 2021](https://yomu.fyi/post/supabase-beta-march-2021.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Apr 6, 2021 Supabase announced a series of platform updates for March 2021, headlined by the launch of Supabase Storage and new local development capabilities. The new storage offering integrates an open-source Fastify and TypeScript API backed by S3, while direct database connections now bundle PgBouncer for connection pooling. Developers can run the platform locally via the Supabase CLI, utilize Docker tooling for self-hosting, and access the open-sourced Supabase UI component library. Community contributions brought OAuth scopes to Supabase Auth for requesting elevated third-party provider permissions, alongside newly published free and pro pricing tiers. In addition, development is underway on an Elixir-based workflow engine compatible with the Amazon States Language for automated event handling. ### [Supabase Dot Com](https://yomu.fyi/post/supabase-dot-com.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Apr 2, 2021 Supabase announced a migration of its primary website from supabase.io to supabase.com over the upcoming week. User API endpoints will remain unchanged on supabase.co, requiring no configuration adjustments for active applications. The company originally selected supabase.io as a meme-inspired placeholder after releasing a Postgres Realtime engine on GitHub in 2019 and failing to find an available dot com name. Following advice from Y Combinator to secure a dot com address, the founders bypassed domain brokers charging percentage fees and negotiated directly with the domain owner, whose business address was located one hundred meters from the founder's home. The migration updates public website destinations while preserving existing developer infrastructure unaffected. ### [PgBouncer is now available in Supabase](https://yomu.fyi/post/pgbouncer-is-now-available-in-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Angelico de los Reyes - Published: Apr 2, 2021 Serverless JavaScript frameworks and developer tools frequently establish separate database connections during bursty traffic, rapidly exhausting PostgreSQL connection limits in the absence of traditional middleware. To resolve connection surges, Supabase integrated the open-source connection pooler PgBouncer directly onto the PostgreSQL server across all newly created projects. Rather than increasing the total number of connections PostgreSQL can open, PgBouncer recycles open connections and queues excess requests until active connections become available. Developers can manage pooling through the dashboard across Session, Transaction, and Statement modes, with Transaction mode recommended for serverless functions despite disabling session features like prepared statements. The default pool size is initially configured to 15 connections. ### [Workflows are coming to Supabase](https://yomu.fyi/post/workflows-are-coming-to-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Francesco Ceccon - Published: Apr 2, 2021 Supabase is developing Workflows, an Elixir-based orchestration engine designed to coordinate complex serverless tasks and event-driven logic. Standard serverless and database-triggered functions often struggle with delayed execution and queuing without external cron processes. To solve this with native Postgres integration, the new engine adopts the open-source Amazon States Language specification to orchestrate functions across platforms including AWS, GCP, Azure, OpenFaaS, and Postgres itself. Execution states, jobs, queues, and logs are deeply integrated with Postgres using the Oban job processing library, though transient in-memory workers are also supported. The engine responds directly to HTTP calls or database change events delivered through Supabase Realtime. ### [Supabase Launches NFT Marketplace](https://yomu.fyi/post/supabase-launches-nft-marketplace.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Ant Wilson - Published: Apr 1, 2021 Non-fungible tokens frequently suffer from the copy-paste problem, where public URLs allow unauthorized users to download underlying media directly from web hosts or IPFS. To address this issue, BuyMeth proposes combining public blurhash thumbnails with full image files encrypted under the active owner's cryptographic key and hosted on IPFS. Completed sales trigger Metamask to re-encrypt the file with the buyer's public key, followed by a one-week escrow challenge window where automated verification matches decrypted image hashes against public thumbnails. Ongoing royalty distributions to previous owners provide a financial mechanism to disincentivize leaking unencrypted original image files. The announced platform serves as an April Fools joke rather than an active Supabase product release. ### [Understanding BigBird's Block Sparse Attention](https://yomu.fyi/post/understanding-bigbird-s-block-sparse-attention.md) - Company: huggingface.co - Author: Vasudev Gupta - Published: Mar 31, 2021 Transformer-based models face severe computational bottlenecks due to quadratic time and memory complexity, making sequences longer than 512 tokens impractical. BigBird addresses this limitation by approximating full attention with block sparse attention, scaling context lengths up to 4096 tokens at a lower computational cost. Instead of attending to all tokens, BigBird combines sliding attention for local context, global tokens for long-range relationships, and random tokens to accelerate information transfer. The model is integrated into HuggingFace Transformers, enabling fine-tuning for tasks like long-document summarization and extractive question answering. When sequences are shorter than 1024 tokens, using original full attention remains recommended. ### [Storage is now available in Supabase](https://yomu.fyi/post/storage-is-now-available-in-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Inian Parameshwaran - Published: Mar 30, 2021 Supabase launched Storage, adding a scalable object store to its existing Postgres, authentication, and API services. Existing open-source storage servers like Ceph, Swift, Minio, and Zenko were evaluated but rejected due to auth incompatibilities and external dependencies like etcd, MongoDB, and Kafka. Consequently, the team implemented a custom Storage API server built with Fastify and TypeScript behind the Kong gateway. Object metadata and access control reside directly in Postgres, leveraging Row Level Security policies written in SQL rather than a proprietary domain-specific language. Objects stream directly to managed backends like AWS S3 using Node streams with minimal in-memory buffering, accompanied by a default one-hour Cache-Control header. ### [Supabase Beta Pricing](https://yomu.fyi/post/supabase-beta-pricing.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Published: Mar 29, 2021 Supabase outlines its beta pricing structure across different service tiers covering database infrastructure, authentication, file storage, and platform operations. Every tier includes a dedicated Postgres database, unlimited API requests, unlimited total users, and access to basic multi-factor authentication and social OAuth providers. Lower tiers offer fixed resources starting at 500 megabytes of database storage and 50,000 monthly active users, with project pausing occurring after one week of inactivity. Higher tiers introduce automated backups, point-in-time recovery, branching billed by the hour, increased egress limits, and compliance options like SOC2, ISO 27001, and HIPAA add-ons. Additional charges apply to resources exceeding tier allocations, including pipeline data processing, log drains, custom domains, and extra concurrent Postgres changes connections. ### [Angels of Supabase](https://yomu.fyi/post/angels-of-supabase.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Paul Copplestone - Published: Mar 25, 2021 Supabase announced a 6 million dollar seed round led by Coatue, Mozilla, and Y Combinator, accompanied by over twenty technical angel investors. To guide its mission of providing an open-source alternative to Firebase, the company deliberately assembled a roster of angel backers comprising CTOs, developers, and developer advocates. The post features short interviews with several angel investors who share their career backgrounds, coding habits, and rationale for supporting the platform. Many investors highlighted the appeal of Supabase's developer ergonomics, its open-source foundation, and the advantages of utilizing Postgres directly alongside built-in authentication over proprietary NoSQL alternatives. Supabase also teased its upcoming Launch Week, during which the team plans to release one new feature daily for a week. ### [Launch week](https://yomu.fyi/post/launch-week.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Ant Wilson - Published: Mar 25, 2021 Following its transition from Alpha to Beta, Supabase scheduled an intensive Launch Week to deliver multiple major features built over a three-month cycle. The event opened with a pricing model designed around predictable metrics like storage to prevent unexpected billing spikes from API requests. The team subsequently launched Supabase Storage, combining a Fastify Node.js API layer with Kong authentication and Postgres Row Level Security for policy evaluation. Additional launches delivered a CLI supporting database migrations and Docker self-hosting, alongside an open-source dashboard UI component library. Finally, Supabase added PgBouncer connection pooling for Postgres instances, previewed an Elixir workflow engine, and transitioned its primary domain to supabase.com. ### [The Partnership: Amazon SageMaker and Hugging Face](https://yomu.fyi/post/the-partnership-amazon-sagemaker-and-hugging-face.md) - Company: huggingface.co - Author: Philipp Schmid - Published: Mar 23, 2021 Hugging Face and Amazon formed a strategic partnership establishing Amazon Web Services as Hugging Face's preferred cloud provider to simplify training and deploying Transformer models. To facilitate this collaboration, the teams introduced Hugging Face Deep Learning Containers optimized for PyTorch and TensorFlow in Amazon SageMaker. The integration provides an extension to the SageMaker Python SDK, enabling data science teams to configure distributed training jobs across single-node and multi-node GPU clusters. Workflows integrate directly with SageMaker Automatic Model Tuning, SageMaker Studio, and Amazon S3 for managed experiment tracking and artifact storage. The open-source containers allow trained models to be exported for inference via SageMaker or published directly to the Hugging Face Model Hub while utilizing EC2 Spot instances to reduce compute expenses. ### [Developers stay up to date with intheloop.dev](https://yomu.fyi/post/developers-stay-up-to-date-with-intheloop-dev.md) - Company: [Supabase](https://yomu.fyi/company/supabase.md) - Author: Rory Wilding - Published: Mar 22, 2021 Freelance engineer Kevin Grüneberg created intheloop.dev as an open-source aggregator to help developers track software updates, release notes, changelogs, tweets, and community discussions from a single platform. The initial release provides dedicated coverage for React, Kotlin, and Vite, with plans to introduce user accounts, notifications, and personalized digests. Although Grüneberg previously used Firebase for rapid prototyping, he found relational requirements like count queries, normalization, and foreign keys difficult or impossible to implement. He selected Supabase as the backend to combine fast integration with PostgreSQL capabilities, including SQL queries, constraints, views, and materialized views. This database architecture allows the application to handle structured data relationships efficiently while scaling future feature additions without restructuring core models. ### [How Grab Leveraged Performance Marketing Automation to Improve Conversion Rates by 30%](https://yomu.fyi/post/how-grab-leveraged-performance-marketing-automation-to-improve-convers.md) - Company: [Grab](https://yomu.fyi/company/grab.md) - Author: Sc Ng - Published: Mar 22, 2021 Grab faced operational bottlenecks managing direct-response Google Ads campaigns across thousands of ad groups due to its hyperlocal marketing across Southeast Asian markets. To eliminate the manual burden of tracking and updating ad creatives, the team built CARA, an in-house automation tool deployed on AWS serverless compute. CARA utilizes standardized file naming conventions to map assets to specific campaigns and connects with Google Ads and YouTube APIs to detect and replace low-performing assets. During an experimental rollout across more than 8,000 active ad groups, CARA replaced nearly 2,000 underperforming creatives. The automated asset replacement workflow produced an 18% to 30% increase in clickthrough and conversion rates. [Newer posts](https://yomu.fyi/page/65.md) · [Older posts](https://yomu.fyi/page/67.md)