Loading…
Browse by type
Explainer reads
Explainer posts on Yomu, summarised newest first.
- 1,318 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.
Explainer posts
MLOps vs DevOps: A Practical Guide for Data Scientists and IT Teams
The guide compares DevOps and MLOps, framing both as practices for reliable production systems but distinguishing traditional application code from machine-learning systems involving data and models. DevOps uses source control, automated testing, staging, and CI/CD, while MLOps governs the “holy trinity” of code, data, and models through separate versioning, quality checks, lineage, and access controls. Its central operational difference is Continuous Training: MLOps pipelines can validate model quality, monitor drift, and trigger retraining when data distributions shift, alongside testing model-serving infrastructure and promoting models through a registry. The guide also assigns responsibilities across data scientists, ML engineers, data engineers, and IT operations, and recommends dedicated MLOps for business-critical, frequently retrained models, DevOps for applications without ML, and split pipelines for hybrid products.
Top Data Warehouse Tools For Modern Data Analytics
Data warehouse tools are presented as a strategic choice for analytics and ML teams facing fragmented estates of warehouses, lakes, and standalone ML systems. The guide proposes evaluating platforms across query performance, scalability, data integration, BI connectivity, total cost of ownership, and governance and security, with attention to MPP, columnar storage, and decoupled compute and storage. It contrasts traditional warehouses, data lakes, and lakehouses: warehouses favor structured SQL analytics, lakes offer native-format flexibility but weaker quality and performance guarantees, and lakehouses combine these capabilities using open formats such as Delta Lake and Apache Iceberg. The stated conclusion is that a lakehouse can provide a single governed foundation for SQL, BI, streaming, ML, and AI workloads, while teams should select tools according to workload, scale, budget, and future AI needs.
Agentic AI vs Generative AI: Comparing Autonomy, Workflows, and Use Cases
Agentic AI and generative AI differ primarily in whether a system autonomously pursues a multi-step goal or produces content in response to a prompt. The post defines agentic systems through a perceive-plan-act cycle: agents maintain memory and state, decompose goals, call tools or sub-agents, evaluate conditions, and recover from errors, while generative AI typically performs bounded, reactive inference. It presents workflow examples such as sales follow-up and market-intelligence summarization, showing how APIs connect agents to CRMs, databases, communication platforms, and news services while LLMs provide text generation or reasoning at individual steps. RAG can ground generative outputs in external knowledge, but agentic deployments add operational concerns around repeated inference loops, human oversight, provenance logging, and access controls. The conclusion recommends choosing by task structure: generative AI for single-turn creation or summarization, agentic AI for autonomous coordination, and both together for complex enterprise workflows.
How Ansible® Automation Powers the Uber Corporate Network at a Global Scale
Uber’s corporate network spans offices across six continents and around 5,000 devices, creating a configuration-management problem involving continuous updates, regional variation, and hardware from multiple vendors. The post explains why Uber uses Ansible for on-premises network configuration while Terraform handles cloud provisioning and Puppet manages configuration on running machines. Its Daily Nightly Enforcement workflow backs up each region’s running configuration to a private GitHub repository, gathers device facts, renders vendor-specific golden configurations through inventory variables and Jinja templates, and pushes them overnight, with failures triggering alerts. Inventory groups encode desired state across regions, functions, and operating systems, while custom roles, formatters, cleanup jobs, and occasional Ansible or Python modules address scale and vendor-specific gaps. The conclusion says this automation makes the diverse network manageable, stable, updated, and ready for daily use, although some hardware still requires other portals.
Ready, Set, Build with the NHS Federated Data Platform
The NHS Federated Data Platform (NHS FDP), delivered by NHS England through a consortium led by Palantir, connects operational data held in separate NHS systems while keeping each organisation’s data under local control. Built on Palantir Foundry, its federated tenants use the Canonical Data Model, encryption, purpose-based access, and Privacy Enhancing Technology for national data flows. As of end December 2025, the post reports 110,078 additional theatre procedures, 797,728 patients safely removed from waiting lists, and 297,943 safely discharged patients, alongside around a 15% reduction in observed discharge delays. The inaugural Build with FDP hackathon brought together over 120 NHS developers and analysts, who built 25 working prototypes on a notional Canonical Data Model to address frontline operational challenges. Projects included early-warning, bed-demand prediction, CAMHS demand balancing, and genomic-optimised waitlist tools.
Validating Bounding Box Annotations
Video bounding-box annotations can contain ID swaps and position jumps that degrade training data for object-tracking models, especially when long videos are split among annotators and later merged. Uber's system runs in uLabel, combining visual, motion, and coordinate features from an 11-frame sliding window with an XGBoost classifier trained on synthetic anomalies from six open-source datasets. It scores each frame for error probability, then groups predictions with frame-index and confidence clustering into regions that operators can review, fix, or dismiss. The current iteration prioritizes jump and ID-swap detection, while freeze and scale/aspect errors remain planned for future releases; the deployed workflow is intended to reduce the cost and time of sequential edit-review while processing client data statelessly without retaining it or using it to train models.
Continually improving our agent harness
Cursor describes its agent harness as an iterative product process: vision-driven hypotheses are tested through offline evals, real-world A/B experiments, and instrumentation. Its context strategy has shifted from static codebase context and guardrails toward dynamic retrieval as models improve, while model-specific prompts and tools match training conventions. Quality measurement combines CursorBench with latency, token efficiency, tool-call errors, cache behavior, code Keep Rate, and language-model analysis of user responses. The team classifies expected and unknown tool errors, uses per-tool and per-model anomaly baselines, and automates log investigations and backlog tickets; one sprint reduced unexpected tool-call errors by an order of magnitude. The harness also supports model switching and subagents, although switching can cause cache misses and summary information loss, and the post argues that multi-agent orchestration will make harness engineering increasingly central.
Backstage with Lakebase
Backstage with Lakebase describes a proof of concept that moves Backstage’s stateful internal developer portal from standard Postgres to Databricks Lakebase, a serverless Postgres surface in the Databricks Data + AI Platform. Connection required an app-config.yaml change, PgSearchEngine instead of the default in-memory search, and short-lived OAuth JWT credentials generated with databricks postgres generate-database-credential; Knex migrations then ran cleanly. Using copy-on-write branches, the team cloned a roughly 63 MB catalog in 1.09 seconds and recovered 32 deleted entities in 3.78 seconds, while observing that point-in-time recovery snaps backward to the nearest WAL record. The post argues that cheap, isolated database state can replace much mocking and staging iteration with real-data development, branch-based testing, schema validation in CI, and automated cleanup, while noting that VS Code/Cursor synchronization tooling remains underway.
Shelf availability starts with better demand visibility
Retail out-of-stock rates in grocery and general merchandise typically run between 7% and 10%, leaving roughly one in ten sought-after items unavailable at a given moment. The revenue impact is real, but repeated shelf gaps can cause customers to build shopping habits elsewhere. Modern retail replenishment requires real-time synthesis of POS velocity by store and SKU, distribution-center inventory, on-order quantities and delivery windows, supplier fill-rate histories, promotional calendars, and other demand signals. Databricks Genie is presented as a conversational interface to the full inventory and demand environment, allowing leaders to ask which high-velocity SKUs are projected to stock out within 72 hours and receive current on-order positions in seconds rather than hours. Its stated differentiators include store-SKU granularity, shared promotional, event, and weather data, supplier performance context, and cross-distribution-center rebalancing; the post says Genie is available today.
Data Mesh at Grab (Part II): The foundational tools behind certification
Grab’s Signals Marketplace uses three foundational platforms to make data certification operational across its data mesh: Hubble, Genchi, and the Data Contract Registry. Hubble, built on DataHub, catalogs assets, ownership, documentation, lineage, contracts, and certification status, while its event-driven engine recalculates certification as metadata and upstream health change. Genchi provides self-service observability and tests for freshness, volume, schema, completeness, and business checks; the registry manages explicit expectations between producers and consumers. Together, these signals let Hubble classify assets as Uncertified, Certified, CertifiedPlus, or Revoked, giving consumers a workflow centered on certified data rather than hearsay. The post reports that P80 datasets, the tables accounting for 80% of queries, fell by over 58% in one year after the Signals Marketplace campaign began in 2024, and positions the foundation as support for AI agents and automated analytics.
Companies winning with AI built the data layer first
Trinity Industries’ experience is presented as a case for treating the data layer, rather than models or dashboards, as the foundation of enterprise AI. The railcar manufacturer migrated 95% of its enterprise data to a single Databricks lakehouse, adopted Medallion architecture, moved transformations upstream, and retired legacy dashboards containing nearly 600 measures. That consolidation supports streaming ETA predictions, procurement agents, and Genie conversational analytics: the ETA model is reported as 50% more accurate than industry ETAs, agents helped increase on-time material delivery by 15%, and Genie handles more than 1,000 questions monthly. The migration took close to a year, followed by six to eight months of additional work, but the account argues that trusted, standardized data enables experimentation, automation, and broader employee access to analysis.
Granite 4.1 LLMs: How They’re Built
Granite 4.1 is a family of dense decoder-only language models in 3B, 8B, and 30B sizes, trained on approximately 15T tokens. Its five-phase pre-training pipeline moves from broad web-scale data to progressively curated mixtures, increases math and code emphasis, and extends context from 4K to 512K tokens through staged long-context training. Supervised fine-tuning uses about 4.1M curated samples evaluated with LLM-as-Judge and rule-based checks, while reinforcement learning uses on-policy GRPO with DAPO loss. The 8B instruct model matches or surpasses the earlier Granite 4.0-H-Small 32B-A9B MoE model despite its simpler dense architecture and fewer parameters, and the Granite 4.1 models are released under Apache 2.0.
Agents are ready, but your architecture probably isn't
Enterprise AI initiatives often produce activity rather than value because organizations start with technology instead of a defined outcome and overlook data architecture, governance, and semantic context. Agentic systems add risk when they can send messages, update records, place orders, or delete records, making permissions and situational controls essential. The discussion argues that dashboards and batch pipelines are poorly matched to low-latency, high-scale agent workloads, which require transactional infrastructure alongside existing analytics. Lakebase is presented as that transactional foundation, while AgentBricks, Databricks Apps, and Genie provide agent development and monitoring, application delivery, and conversational data access. The recommended path is to define success first, isolate a focused pilot, learn what works, and redesign underlying processes rather than merely add AI to them.
Building an RL theorem-proving workflow on Modal
AE Studio built a distributed reinforcement learning pipeline on Modal to train LLMs to prove theorems in Lean. They compared Group Relative Policy Optimization, which groups attempts and applies gradient updates, with Evolution Strategies, which evaluates a population of perturbed models and updates based on success rates. The workflow used Modal’s per-function images to separate a vLLM instance on GPU for proof generation, CPU-based Lean verification, and orchestration, with .map() fanning out runs and Sandboxes isolating verification batches. ES enabled stateless checkpointing via seed histories: each worker loaded base weights from a Volume, replayed past seeds and rewards (about 200 bytes per iteration), and reconstructed the model on GPU without transferring large weight files. Early results on the lean_workbook dataset using the Kimina-Prover-Preview-Distill-1.5B model showed ES matching or outperforming GRPO in verified proofs per iteration, especially with limited data, though gains varied.
Built in, not bolted on: What AI-native actually means in cybersecurity
The discussion defines AI-native cybersecurity applications as systems architected with intelligence at their core, rather than traditional products with AI added later. It links tool sprawl to slower threat detection and response and argues that proprietary, context-rich security telemetry is the foundation for adaptive defense. Barracuda uses Databricks to consolidate fragmented data, normalize schemas, support real-time streaming detection, run ML operations through MLflow, and power natural-language log search across billions of security events with strict data isolation. The work began by defining customer outcomes, then progressed through incremental delivery; normalized data enabled models and agents to use cross-domain context. This approach extended across WAF-as-a-service, automated configuration, API security, and bot protection while shared outcomes aligned product, data science, engineering, and business teams.
Using Uber at Louis Armstrong New Orleans International Airport
Louis Armstrong New Orleans International Airport (MSY) handled more than 10 million passengers in 2015, a nearly 9% increase from the previous year, making preparation useful for travelers. The post recommends Uber as an alternative to parking, with uberX or uberXL available and pickup at the Ground Transportation Center on the first floor of the short-term parking garage. After collecting luggage and selecting a vehicle, riders should follow the airport signs; drivers wait at designated parking spots, and riders can call if they have trouble connecting. Additional resources cover terminal maps, traffic, flight status, airport updates, TSA guidelines, and security wait times. It also points visitors toward the French Quarter, Jean Lafitte National Historical Park and Preserve, and Casamento’s, noting attractions, trails, historical sites, gumbo, and an oyster loaf sandwich.
Where to Celebrate Valentine's Day in Pittsburgh
Pittsburgh’s Valentine’s Day guide pairs restaurant suggestions with movies, concerts, creative courses, or a cooking class, allowing couples to mix and match stops. Dinner options include Dish Osteria and Bar’s daily-changing classic Italian menu, Morcilla’s shareable Spanish pintxos, Kaya’s Caribbean-inspired dishes, and Everyday Noodles’ soups, dim sum, and soup dumplings. Events range from films at the Manor Theatre and February concerts at Mr. Smalls Funhouse and Club Cafe to multi-week courses at Manchester Craftsmen’s Guild and a February 11 class at Habitat, which includes lunch and an apron. The practical notes include making reservations at Morcilla, arriving early for a Club Cafe table, and scheduling Uber between destinations for convenience.
New Atlanta Stadium ride guide: Getting there
Uber’s guide explains transportation options for events at Atlanta’s new Mercedes-Benz Stadium, which it says will host more than 100 events that year. It recommends combining Uber with MARTA: riders can use promo code MARTA2MBS for 50% off eligible uberX or uberPOOL trips to and from select stations, within specified event windows. The discount has exclusions, a maximum value of $10, and applies to up to five rides per event in select Atlanta areas while supplies last; the guide lists dates and times ranging from September 7 through January 8. For direct stadium trips, Uber advises using designated pickup points, avoiding Northside Drive, verifying the vehicle and license plate, and using features such as upfront fares and Split Fare.
Perfect Date Ideas in Manhattan for Valentine's Day
Manhattan is presented as a Valentine’s Day destination for couples seeking either elaborate experiences or quieter, indulgent ways to spend the evening. The recommendations range from a Hornblower Cruise’s New York City Lights Dinner at Pier 40, with filet mignon and landmark views, to dancing at the vintage Jane Ballroom and rooftop views across the Hudson. For a celestial theme, Hayden Planetarium offers a hand-holding setting where Neil deGrasse Tyson guides visitors through astronomy. The Library at the NoMad Hotel pairs Old World décor, books, and more than 250 bottles of bubbly, while AMC Loews on West 84th Street offers reclining seats that convert into love seats for a movie night. Together, the suggestions cover waterfront dining, dancing, stargazing, champagne, and a cozy film.
5 Beaches in Close Proximity to NYC
The article recommends five beach escapes near New York City, combining public transportation with Uber rides to reach destinations beyond much of the city’s less inviting coastline. Orient Beach State Park offers eight miles of Gardiner’s Bay shoreline, undeveloped acreage, nature trails, and views of Shelter Island. Fort Tilden pairs a pristine Queens shoreline with former military structures and art installations, while Robert Moses State Park provides a less crowded Fire Island setting with picnic areas, mini-golf, surfing, and fishing. Jones Beach adds a broad, popular barrier-island destination with sports, a nature center, concerts, and quieter sand beyond its facilities. Sherwood Island State Park in Connecticut supplies a quieter mix of sand, woods, wetlands, picnicking, and salt-marsh trails.
Top topics
- Architecture830
- Performance812
- LLMs648
- Machine Learning617
- Open Source561
- AI Agents542
- AI454
- Developer Experience446
Weekly digest
The best engineering reads, once a week.