What we're learning, shared openly.
A worked AWS to GCP compute migration: priced machine-family choices, Savings Plan wave order, Spot drain rewrites, GKE pod CIDR math, Workload Identity and an Inferentia port, with a cutover and rollback plan.
BigQuery pricing math, six SQL differences that silently change results, Kinesis versus Pub/Sub ordering and cost, and how long to run Redshift and BigQuery side by side before finance switches.
A worked AWS to GCP serverless move: when Cloud Run beats Lambda on cost and when it doesn't, what breaks under concurrency, how SQS maps to Pub/Sub, what Workflows can't do that Step Functions can, and how to cut over by traffic weight.
Moving 400 TB from S3, EBS and EFS to GCP: why object count, not terabytes, sets the bill and the timeline, and how to verify and cut over without losing a file.
Why the GKE provider drops disruption_budget with recurring_maintenance_window, why an empty master_authorized_networks_config never converges, which workaround to pick, and a gcloud and jq gate that catches both.
Shopify moved a production app in 12 weeks with AI agents. Six lessons for any legacy modernization, plus the last mile their post leaves out.
How an orchestrator, parallel porter agents and an independent reviewer can multiply legacy modernization throughput, and what has to be true first.
Layered gates, characterization tests, maker-checker review and visual diffs: how to set up AI agents as a QA team for legacy modernization.
No inventory, big-bang rewrites, ports without tests, agents with no gates and seven more modernization mistakes, with what to do instead.
PII Extractor-68M scores 88.3% F1 on Privacy-Bench, ahead of PII-Tracer (86.5%) and GLiNER2-PII (83.7%), with 37% fewer false alarms than PII-Tracer. Three open models, from 17M to 68M parameters.
Which AWS service replaces each GCP compute option, what works differently, what changes on the bill, and how to do the move step by step.
Which AWS service replaces BigQuery, Dataproc, Dataflow and Pub/Sub, what works differently, what changes on the bill, and how to do the move step by step.
Which AWS service replaces Cloud Run, Cloud Run functions, App Engine and Eventarc, what works differently, what changes on the bill, and how to do the move step by step.
Which AWS service replaces Cloud Storage, Persistent Disk and Filestore, how storage classes and access change, how to copy the data, and what changes on the bill.
A bug hiding in PHP since 2012 lets anyone run code on your server, no login needed. How it works, how it was found, and how to fix it.
A static field is shared by every user of the server. Before you can move a Java web app, you have to find out what lives in memory, because that decides what has to move together.
LLM judges re-read the whole rubric on every call. RubriX reads each level once and scores every response in a single pass: about 28 ms per judgment from 2 to 100 levels, up to 7.7× faster than Jev, and 75.4% exact accuracy on held-out rubrics against 57.1% for Jev.
In COBOL, the copybook is the real contract. Get the layout or the arithmetic slightly wrong and nothing crashes. The totals are just off by a cent.
Your API clients depend on PHP's loose rules, whether you meant them to or not. Rust makes you write every one of them down.
Big-bang rewrite, strangler fig, branch by abstraction or vertical slices? How each migration strategy works, when to use it, and the combined playbook we follow.
Wrap legacy events in Observables, keep shared state behind a stream that both apps subscribe to, and replace hand-written debounce, retry and polling code with a few operators.
MVC, MVVM and one-way data flow all answer the same question: where does state live, and who is allowed to change it?
Two-way and one-way binding each have a job. Here's how both work, and how to migrate forms from Ember or Angular to React, and from React to Angular or Vue, without breaking them.
CSR and SSR each have a job. Here's how both work, and how to move a React SPA to server rendering, or a server-rendered app back to a SPA, without breaking it.
There's no framework to leave behind in a vanilla JavaScript app. What you do have is state hiding all over the page, and finding it is most of the job.
In a jQuery app, nothing updates the total by itself. Every handler has to remember to do it, and the bugs live in the ones that forget.
You aren't really migrating off Backbone. You're migrating off the framework your team built on top of it, and most of that can simply be deleted.
Most Angular-to-React migrations don't fail on a tricky hook or a missing library. They fail on a decision made in the first week. Here's what we learned porting large applications from Angular to React.
Handlebars to JSX is the easy part. The real work in an Ember-to-React migration is everything Ember was quietly doing for you.
We ran TypeSafe's Jev as a prompt guardrail on 1,580 red-teaming prompts against GPT-5-mini, Claude Haiku 4.5 and three open-source guard models. It was five to six times faster, 6–27× cheaper, and had the highest block accuracy of the hosted models.
AI can meaningfully improve collections performance within the regulatory framework: propensity scoring, channel optimisation, timing models. What it cannot do is automate the things TCPA, FDCPA, and state-level consumer protection laws prohibit. Here is what the boundary looks like.
Better models have not solved the thin-file problem, the data quality issues in alternative data, or the validation requirements that govern what can be deployed in regulated lending. The data problems are still the hard part.
Rules-based fraud detection is fast to deploy and easy to explain. It also has a ceiling: fraudsters adapt to rules, false positive rates are hard to reduce, and new fraud patterns take time to codify. ML adds a measurable lift above that ceiling: here is what that looks like in practice.
Intelligent document processing handles pay stubs, bank statements, and tax returns well under controlled conditions. The 20% exception rate, downstream system integration, and confidence threshold design are where most production deployments get complicated.
Lenders deploying AI in underwriting face a compliance landscape that shapes what models can be used, what documentation is required, and how adverse action notices must work. Here is what the regulatory framework actually requires and what you can deploy today.
High-mix low-volume manufacturing scheduling is one of the harder operational problems in industry. ERP scheduling modules are built for repetitive production. Custom rules break under complexity. Here is what AI scheduling actually delivers and what it does not.
Predictive maintenance pilots succeed regularly. Production deployments are harder. The difference is usually not the model: it is asset coverage, alert fatigue, maintenance workflow integration, and the trust gap that builds when the first false positives arrive.
Vision-based AI inspection is mature for specific defect types on uniform parts. Process parameter correlation is underused. Root cause analysis and real-time process control are different problems requiring different architectures.
Most modern plants log process parameters every second, store years of it in historians, and act on almost none of it. The gap is not data collection: it is the connection between what is stored and a decision someone can act on.
ERP demand forecasts look reasonable at the product family level and fall apart at the SKU level: the level where production scheduling, material purchasing, and safety stock decisions actually get made.
GPT-4 scores 87% on HumanEval. Claude solves competitive programming problems. Yet engineering teams are merging regressions, leaking secrets, and shipping logic bugs they did not write. What the benchmarks are measuring and what actually ships are two different things.
Runpod is basically a cloud platform for AI and ML work. What makes it unique is GPU compute on-demand - like Uber for GPUs. Pay only for what you use, without owning or managing infrastructure.
Resource planning in most manufacturing organizations remains a juggling act. Planners rely on disconnected spreadsheets stitched together from Enterprise Resource Planning (ERP) systems...