Loading…
Browse by type
Problem & solution reads
Problem & solution posts on Yomu, summarised newest first.
- 934 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.
Problem & solution posts
Reclaiming Terabytes: Optimizing Android image caching with TLRU
Grab’s Android app cached runtime images with DiskLruCache, but its 100 MB size limit did not remove stale content while the cache remained below capacity, contributing substantially to device storage usage. The team developed Time-Aware Least Recently Used (TLRU), which combines LRU size eviction with time-based expiration using last_accessed metadata and a TTL, while a minimum cache-size threshold protects a baseline of cached images. TLRU retains the 100 MB maximum and evicts entries when current_time minus last_accessed exceeds the TTL; the implementation also supports reading existing LRU journal files by ignoring timestamp suffixes, enabling safe rollback. Configuration was tuned through experiments using cache hit ratio, with a maximum acceptable decline of three percentage points. After production rollout, P95 total app size fell approximately 50 MB, while hit ratio stayed within target and infrastructure costs did not increase.
PlanetScale protects production reliability with Bugbot
PlanetScale describes code review as the bottleneck created by rapidly increasing agent-generated code, while human review capacity remained fixed. To preserve production reliability, the company adopted Bugbot as a dedicated agentic review layer alongside its existing process. Bugbot surfaces semantic and logical issues that static analyzers, linters, and straightforward frontier-model reviews may miss, including premature completion states, non-converging asynchronous controllers, and database restart edge cases. Across more than 2,000 pull requests reviewed monthly, roughly 80% of Bugbot's comments are addressed before merge. PlanetScale says this prevents potential production downtime, saves the equivalent of two full-time engineers of review effort, and lets engineers focus on complex infrastructure work without sacrificing quality.
Towards Model-based Verification of a Key-Value Storage Engine
MongoDB developers formalized the interface boundary between their distributed transactions protocol and the underlying WiredTiger key-value storage engine using a compositional TLA+ specification. This modular model established an abstract contract to verify that the storage engine's concrete timestamp-based operations matched the semantics expected by the distributed transaction layer. To check conformance, a modified version of the TLC model checker generates a complete graph of reachable states for finite parameters and computes path coverings. Each identified path is translated into an automated test case consisting of a sequence of storage engine API calls. For a finite model with two keys and two transactions, this workflow generated 87,143 test cases that executed against WiredTiger in approximately 40 minutes.
Innovating with MongoDB | Customer Successes, February 2026
Modern artificial intelligence startups frequently face operational bottlenecks when deploying autonomous agents and dynamic interfaces on rigid relational schemas or restricted database engines. To address these limitations, platforms including Modelence, Thesys, Emergent Labs, and Heidi integrated MongoDB Atlas as their operational data foundation. Emergent Labs eliminated recursive schema migration failures previously encountered with PostgreSQL by adopting document structures that match native JSON generation, supporting millions of deployed applications. Meanwhile, Heidi migrated from Amazon DocumentDB to utilize MongoDB Vector Search for retrieval-augmented generation pipelines under a single unified API without requiring external search databases. These architectural transitions streamlined continuous agent iterations, eliminated operational scaling downtime, and accelerated deployment cycles across diverse production environments.
Scaling Localization with AI at Lyft
Lyft re-architected its batch localization pipeline to combine artificial intelligence with professional linguist oversight. The previous workflow relied exclusively on human translation, which resulted in multi-day turnarounds and linear cost scaling that hindered rapid market expansions. Under the new architecture, a dual-path system generates three candidate translations using a fast generation model, evaluates them against defined rubrics using a reasoning-focused model, and immediately releases approved drafts. Linguists review the drafts asynchronously in the Smartling translation management system to establish the final system of record. This pipeline reduced delivery latency from days to minutes while achieving a state where 95% of machine-generated translations require no significant changes during human review.
Implementing a secure sandbox for local agents
Auto-approved terminal commands make coding agents more capable but increase the risk of destructive changes and secret leakage, while repeated approval prompts create approval fatigue. Over three months, the team rolled out a uniform sandbox API across macOS, Linux, and Windows, allowing agents to run freely inside controlled environments and request approval mainly for internet access. macOS uses dynamically generated Seatbelt profiles; Linux combines Landlock, seccomp, and overlay filesystems to enforce workspace and ignored-file restrictions; Windows runs the Linux sandbox through WSL2. Changes to tool descriptions and result rendering help agents recognize constraints and escalate instead of retrying failed commands. Production data reports that sandboxed agents stop 40% less often, and the sandbox now handles a third of requests on supported platforms, although native Windows isolation remains under development.
Trusting the Untestable: Validation and Diagnostics for the Doubly Robust Models
When randomized A/B tests are infeasible for evaluating partnerships or long-term effects, Lyft applies Augmented Inverse Propensity Weighting (AIPW) doubly robust models to estimate treatment effects from observational data. The quasi-experimentation platform mandates rigorous confounder selection prior to treatment exposure and adjusts for downsampling bias via propensity score conversions and outcome reweighting. When comparing AIPW against experimental ground truths, initial estimates understated effects because trimming extreme propensity scores removed key users and unobserved confounders persisted. To evaluate model health and external validity, the platform incorporates diagnostic scorecards featuring common support checks, Marginal Sensitivity Models, and covariate comparisons between trimmed and untrimmed cohorts.
Balancing cost and reliability for Spark on Kubernetes
Notion describes how it reduced Spark compute costs on Kubernetes while maintaining reliability across production workloads. After moving from fixed EC2 clusters on Amazon EMR to EMR on EKS, it used Karpenter, EKS Auto Mode, and the MostAllocated scheduler for dynamic provisioning and bin-packing. Spot Instances then exposed a failure mode: one interruption could remove many bin-packed executors, causing shuffle loss and expensive recomputation. Spot Balancer addresses this by labeling executor creation through a Kubernetes webhook and maintaining each job’s configured spot-to-on-demand ratio, with per-job counts stored in Redis. Named stability weights make the cost-reliability trade-off explicit, and reported savings reached 60–90 percent while jobs that previously failed from interruptions finished reliably.
Litestream Writable VFS
Litestream’s VFS is being extended to support Sprites, whose storage stack needs SQLite-backed block maps to serve writes immediately during cold starts, before a full database or block map is available. The VFS indexes each database page by file, offset, and size in LTX files, serves reads from object storage, and now optionally buffers writes locally before syncing them roughly every second or on clean shutdown. Write mode assumes a single writer, disables polling, and offers only the same eventual durability already provided by Sprite storage. A separate hydration mode serves remote reads while reconstructing a complete database file in the background, then switches reads to that file; the temporary file is discarded on exit. These features target Sprite-like workloads, while ordinary Litestream sidecar deployments do not require them.
We Got Claude to Build CUDA Kernels and teach open models!
Domain-specific coding tasks like writing optimized CUDA kernels often require capable, expensive frontier models, leading to high operational costs. The upskill tool addresses this challenge by capturing specialized workflows from teacher models such as Claude Opus 4.5 into portable skill files that can be consumed by smaller or locally hosted models. The framework automatically generates input-output test cases from execution traces to benchmark student model accuracy and token efficiency with and without the skill. In an evaluation targeting HuggingFace kernel-builder development, a local GLM-4.7 Flash model improved its pass rate from 40% to 85% when supplied with the generated skill. Measuring token usage across different models remains essential, as skills can lower token consumption for some architectures while introducing unnecessary overhead for others.
Securely indexing large codebases
Cursor describes a secure way to index large codebases for semantic search, where naive processing can delay availability for hours. The system builds a Merkle tree of file and directory hashes, synchronizes only divergent branches, and caches embeddings by chunk content to reduce repeated work. For new users, a client-generated simhash finds a similar index within the same team or user's indexes, which is copied while background synchronization reconciles differences. Content proofs derived from the client's full Merkle tree filter search results so users see only files they can prove they possess. Reported time-to-first-query falls from 7.87 seconds to 525 milliseconds for the median repository, and from 4.03 hours to 21 seconds at the 99th percentile.
Unlocking Agentic RL Training for GPT-OSS: A Practical Retrospective
Engineers evaluated the GPT-OSS model for agentic reinforcement learning across multi-step tasks such as ReTool and GSM8K using the verl framework. Initial training runs suffered from exploding gradient norms, increasing KL divergence, and stagnating rewards. Investigation revealed that non-deterministic expert routing in the Mixture of Experts architecture caused log-probability mismatches between forward passes, which falsely triggered Proximal Policy Optimization clipping. The team resolved this by substituting detached current log-probabilities for old log-probabilities during on-policy updates. To address further training instabilities and memory bottlenecks, the implementation integrated an attention-sink-aware backward pass into FlashAttention v3 and applied sequence parallelism using all-to-all communication around attention layers.
Salesforce accelerates velocity by over 30% and ships higher-quality code with Cursor
Salesforce has made Cursor available to thousands of engineers working across a codebase built over more than 25 years, and reports more than 30% higher velocity alongside gains in code quality. Junior engineers used it to understand existing code and contribute more effectively, while senior engineers began with tedious tasks before expanding to higher-value work as trust grew. Adoption spread from small groups that demonstrated impact, and AI eventually reached every part of the software development lifecycle rather than code writing alone. Salesforce tracks cycle time, quality measured by bug count, and throughput; it reports double-digit improvements across all three, an 85% reduction in one team’s legacy code coverage time, and far more unit tests generated with Cursor.
Docker lazy loading at Grab: Accelerating container startup times
Large Docker images for Airflow and Spark Connect caused minute-scale downloads, delaying cold starts and weakening autoscaling responsiveness. The post describes testing eStargz and Seekable OCI (SOCI) with containerd remote snapshotters, which expose registry-backed layers through a virtual lowerdir and fetch file content on demand. In production, SOCI delivered 30–40% faster startup times for Airflow and Spark Connect, while benchmark results showed SOCI matching standard images’ 5.0-second Airflow application startup; eStargz took 25.0 seconds. Configuration tuning increased download concurrency, unpack concurrency, and chunk size, reducing fresh-node image download time from 60 seconds to 24, a 60% improvement. Grab chose SOCI because Bottlerocket supports it natively, while noting that lazy loading redistributes rather than removes download work and requires workload-specific tuning.
From deployment slop to production reality: How BriX bridges the gap with enterprise-grade AI infrastructure
BriX addresses “deployment slop,” the gap between quickly built AI prototypes and enterprise production, where inconsistent versions, reliability failures, and security risks can block adoption. The platform centralizes production infrastructure so semi-technical domain experts can configure deployments rather than become DevOps specialists, covering scaling, monitoring, audit trails, access controls, data connections, and consistent behavior. BriX supports switching among models such as Claude, GPT, and Gemini; its Model Context Protocols (MCPs) provide standardized, pre-built connectors to approved enterprise data sources with built-in access controls and query logging. Centralized system prompts and context files enforce shared guardrails, while custom frontends and persistent, shareable reports extend use beyond chat interfaces. The stated result is a platform intended to turn deployment into a configuration task, helping teams move from prototypes to enterprise-grade tools in days rather than months.
How Merge brought together all it knows using Notion AI
Merge found that company knowledge was scattered across Slack, Google Docs, Salesforce, and other tools, causing repeated questions, slow onboarding, and time lost searching for context. It used Notion AI as a shared knowledge layer, connected Salesforce and Slack, and gave teams guidance on what content to add, including detailed project documents, meeting transcripts, and structured databases. Teams populated the workspace with content calendars, sales enablement materials, hiring plans, roadmaps, and an AI Prompt Library, while all-hands and team meetings demonstrated practical workflows. Merge reports that employees now ask Notion AI first, saving each team member several hours per week, helping teams make decisions with less manual searching, and making faster information retrieval one of the most-cited contributors to improved satisfaction in its latest company-wide survey.
Demystifying user journeys: Revolutionizing troubleshooting with auto tracking
Troubleshooting critical issues in the Grab app was hindered by incomplete user-journey data, because manually instrumenting hundreds of UI components left interaction gaps and delayed diagnosis. AutoTrack is a new SDK that records application state, user interactions, and UI screens in real time, creating an end-to-end session chronicle from launch through navigation. It tracks lifecycle transitions and launch mechanisms such as deeplinks and push notifications, while UIViewController, Activity, and Fragment lifecycle events identify the displayed screen. Screen metadata includes the full hierarchy as JSON; after personal information is redacted for GDPR/PDPA compliance, that data is used to train an AI model that generates an HTML representation of the screen and interactions. AutoTrack supports reproducing elusive bugs, automated issue assignment, UI test generation, and analytics-trigger validation, with future work adding network and performance signals, privacy guardrails, and automated triage.
Token-count-based Batching: Faster, Cheaper Embedding Inference for Queries
Serving embedding models for short search queries often suffers from poor GPU efficiency because traffic is spiky and memory-bound. Traditional time-window and request-count batching strategies lead to inconsistent GPU utilization, while tensor padding wastes compute on empty tokens. To resolve these bottlenecks, Voyage AI implemented token-count-based batching paired with padding removal in inference engines like vLLM. The architecture uses Redis with Lua scripts to atomically aggregate pending requests until reaching an optimal hardware saturation token threshold. Across production deployments, this approach achieved a 50% reduction in GPU inference latency with 3X fewer GPUs and improved throughput by up to 8×.
How Grab is accelerating growth with real-time personalization using Customer Data Platform scenarios
Grab’s Customer Data Platform (CDP) introduced Scenarios to address the limits of daily batch attributes for real-time personalization across marketing and user journeys. The capability combines Scribe event triggers, geo-fencing, optional historical CDP data, and predictive models, with a self-serve GUI for configuring and validating event sequences and APIs for programmatic management. Incoming events are processed through Flink, with historical data evaluated in StarRocks and outputs written to Kafka, Amphawa, or both for downstream consumption; configured Scenarios become available within an hour and are designed for latency under 15 seconds. More than 12 implementations are live, including a Grab Unlimited abandonment intervention that notified users 15 minutes after signup drop-off and produced over a 3% uplift in subscriber conversions versus non-real-time acquisition campaigns. Planned work includes scalable Scenario sharding, multiple output destinations, and delayed triggering.
PaaS + IaaS: Heroku and AWS for Cloud-Native Enterprise Agility
Modern enterprises can combine AWS IaaS infrastructure scale with Heroku’s managed PaaS to reduce operational work and developer cognitive load while retaining control for specialized workloads. Heroku supports Git-based deployment, managed services including Apache Kafka and Heroku Postgres, polyglot applications, and isolated environments such as Private Spaces and Shield Private Spaces. In Salesforce’s example, five applications serving more than 76,000 employees used Heroku to cut expense approval time from two days to two hours and recover $2 million in unused licenses. The architecture also combines Heroku AppLink, Heroku Connect, Heroku Fir’s OpenTelemetry support, and managed Dynos, while AWS remains the stated choice for highly specialized AI training requiring raw GPU access.
Top topics
- Architecture834
- Performance817
- LLMs673
- Machine Learning622
- Open Source568
- AI Agents564
- AI483
- Developer Experience448
Weekly digest
The best engineering reads, once a week.