标签：#Serverless

I've been building AI agents for a while now, and streaming responses to a UI has always been the painful part. In previous projects I tried API Gateway streaming, Lambda response streaming, and even AppSync Events via an agent tool call to notify the UI. I also looked at adding my own WebSocket API through API Gateway, which requires managing $connect , $disconnect , and $default routes, storing connection IDs in DynamoDB, and posting messages back through @connections . All of these approaches felt like too much ceremony for what should be simple. That's when I found that AgentCore has built-in WebSocket support. The browser can just connect directly to the agent. No middleman. I built WearCast to demonstrate this functionality. This is an AI agent that helps you pick out what to where based on the weather. I've found this very helpfu when packing for a trip. The code for this application is public and you can find the full implementation on this GitHub repo . The problem with the traditional approach We usually build app applications as we do REST APIs User → REST API → Lambda → AI Service → Lambda → REST API → User Every message makes a round trip through multiple intermediaries. The response waits until the entire generation is complete, and then it all comes back at once. For a chat interface, this feels sluggish. Users stare at a spinner while the model generates hundreds of tokens they could already be reading. Even if you add Server-Sent Events or long-polling, you're still stitching together a real-time experience on top of infrastructure that seemed like overkill. I wanted something better. The architecture Here's what I ended up with: ┌─────────────┐ ┌──────────────────┐ │ React UI │─── JWT ─▶│ API Gateway + │──▶ Lambda (presigned URL) │ (Cognito) │ │ Cognito Auth │ └──────┬──────┘ └──────────────────┘ │ │ WebSocket (SigV4 presigned URL) │ ← No middleman! Direct connection → ▼ ┌──────────────────────────────────┐ ┌────────────────────┐ │ AgentCore Runtim

Database branching is one of the best ideas serverless Postgres brought to the mainstream. Fork the database at a point in time, get an isolated copy with all the data, run something risky against it, throw it away. It made preview databases and safe migrations feel routine. But a real application is not just a database. It is a database, plus the files it stores in object storage, plus the backend code that serves it, plus, increasingly, the model and gateway config it calls for AI. When you branch only the database, those other three stay shared. Your "branch" points at the same S3 bucket, the same deployed backend, and the same AI configuration as everything else. So it is half a copy, and the half it leaves out is where a lot of the interesting bugs and the scary migrations live. Neon's platform preview changes what a branch contains. A branch now forks the database and its data, the object storage and its files, the functions that run your backend, and the AI gateway config, all at the same point in time, all isolated. A branch stops being a database copy and becomes a whole environment. To make sure that is a real claim and not a diagram, I took a full-stack project, branched it, and checked every layer. Here is what happened. TL;DR Elsewhere, "branch" means the database only. Object storage, backend deploys, and AI config stay shared, so you bolt on scripts to fake per-branch versions of them. A Neon branch forks all four together: Postgres + data, object storage + files, functions (each branch gets its own URL), and the AI gateway. I proved it: branched a project with a DB, a bucket of files, a function, and the gateway. The branch came up with a copy of the rows, a copy of the files on its own storage endpoint, its own function URL, and the gateway. A write to the branch left main untouched, and deleting the branch removed all of it. That makes a branch a real environment: true preview stacks, whole-state bug reproduction, and disposable sandboxes for agent

At re:Invent 2025, AWS announced Lambda Durable Functions . The feature introduces a checkpoint/replay mechanism that allows Lambda executions to run for up to one year, automatically recovering from interruptions by replaying from the last checkpoint. Lambda's 15-minute timeout is not a bug or a limitation to work around. It is a deliberate design choice that encourages keeping functions simple and focused, and in most cases it does its job well. When a function needs more time, the usual approach is fanout : split the work into smaller Lambdas, orchestrate them, move on. I have done it many times and it works perfectly fine. But a few days ago I was developing a new Lambda function for a pipeline orchestrated by Step Functions, and the execution time exceeded 15 minutes. I could have done the usual split, but durable functions had just come out and I wanted to try them. At first glance, durable functions can look like a replacement for Step Functions. Both services manage multi-step workflows , both offer checkpointing and automatic recovery , and both let you coordinate complex operations. For certain use cases, that might actually be the case: if your entire workflow lives inside a single Lambda, durable functions can handle everything on their own without an external orchestrator. But the AWS documentation actually suggests using them together. The "Hybrid architectures" section says it explicitly: many applications benefit from combining the two services, using durable functions for application-level logic within Lambda and Step Functions to coordinate the high-level workflow across multiple AWS services. My case fit that description, and more than a perfect architectural match, I wanted to learn how the two services actually work together and form my own opinion on when the hybrid approach makes sense. I figured integrating the two would be a small change. It was my first time working with the durable execution SDK, and since the code I write is mostly infras

I built a tool that answers one question: when a website won't load, is it your connection or the site? It runs two checks in parallel — measures your own line in the browser (latency + a 1 MiB download), and probes the target from Cloudflare's edge — then returns one of four verdicts: it's not you, it's you, it's both, or all clear. Demo: https://itsnotyou.site The interesting part isn't the tool, it's that the whole thing — app, ~120 monitored sites with 24h history, SSR status pages, share cards, outage alerts — runs for $0/month plus domains. The one thing that wanted to cost money Everything fit the Workers free plan except a background monitor probing ~120 sites on a tight cadence. As a Worker cron that's ~120 subrequests/run and, done naively, thousands of KV writes/day — which pushes you onto Workers Paid. The real ceiling is KV: 1,000 writes/day. So I split it. The user-facing test stays on Cloudflare — the edge probe still measures from the colo nearest the user, which is the point of edge. But the background sweep moved to a cheap VPS I already had: it probes the roster on a systemd timer and pushes results back into KV over the REST API. Staying under 1,000 KV writes/day One KV key, not key-per-site. All ~120 statuses live in a single blob. Key-per-site at sweep cadence would be millions of writes/month; a single blob is one write per sweep. Widen the cadence. I started at 2 min = 720 writes/day. Cloudflare emailed that I'd hit 50% of the daily limit (the sweep wasn't the only writer). I went to 3 min = 480/day, leaving headroom for share cards and the notify list. Move the hot counter off KV. The anonymous "tests today" counter was the sneaky risk — a traffic spike could exhaust the write budget and stall the status sweep, the one thing you can't let happen. It went to Analytics Engine instead (free, uncapped, separate budget). Now no amount of user traffic can starve the monitor. Reader and writer share code so their rules never drift: the streak/histo

When Neon became the official database partner of DEV Community, I was already a user. But the partnership made me look closer at why I chose Neon — and whether those reasons apply to other AI developers. They do. Here's why Neon is the ideal database for AI applications in 2026. The Problem: AI Apps Have Unique Database Needs AI applications have database requirements that traditional web apps don't: High write volume — every AI request generates logs, metrics, and cost data Variable load — traffic spikes when a model goes viral, then drops to zero Schema evolution — you're constantly adding models, routing rules, and analytics tables Dev/prod parity — you need to test routing changes against real production data Edge compatibility — AI APIs need sub-100ms response times globally Traditional PostgreSQL (RDS, Aurora) struggles with all five. Neon was built for them. Feature 1: Database Branching (The Game-Changer) This is Neon's killer feature. It works like git branch but for your entire database: # Create a branch from production neon branches create --parent main --name test-deepseek-v31 # Get a connection string for the branch neon connection-string test-deepseek-v31 # → postgresql://...@ep-test-deepseek...neon.tech/neondb # Run migrations on the branch npx prisma db push --url $BRANCH_URL # Test your new routing algorithm against REAL data # (the branch is a copy-on-write clone of production) # When tests pass, merge neon branches merge test-deepseek-v31 Why This Matters for AI Apps When I added DeepSeek V3.1 to my model pool, I needed to test: Would the new model break existing routing rules? Would the cost calculations be correct? Would the latency meet my SLA? With traditional PostgreSQL, testing against real data meant either: Copying production to a staging DB (hours, $$) Testing with synthetic data (unreliable) With Neon branching, I branched, tested in 30 seconds, and merged. Zero downtime, zero risk. Feature 2: Scale-to-Zero (Cost Optimization) Neon's c

Your order-processing Lambda starts sending duplicate confirmation emails. Not always — maybe one order in twenty. CloudWatch shows more invocations than messages published. The function code hasn't changed in weeks. What changed is that someone added a fraud check that pushed processing time from 25 seconds to around 45, and your SQS queue is still running the default 30-second visibility timeout. That combination is the whole bug. When a Lambda pulls a message from SQS, the message isn't deleted — it's hidden for the duration of the visibility timeout. If the function is still working when that window closes, SQS assumes the consumer died and hands the same message to another invocation. Now two Lambdas are processing the same order, both will "succeed," and both will send the email. Nothing errors. Nothing retries. There is no log line that says "this message was delivered twice because your timeouts are misconfigured." Infrawise ( npm ) flags this exact mismatch as a high-severity finding before it costs you an afternoon of staring at idempotency-free handler code. This post walks through why the bug is so hard to see, how the detection works, and how to keep an AI assistant from reintroducing it. Why you never catch this one yourself Three things make this misconfiguration nearly invisible: It passes every test. In local tests and staging, your handler processes a synthetic message in two seconds. The 30-second visibility window never comes close to expiring. The bug only exists under production conditions — real payload sizes, real downstream latency, cold starts stacking on top of slow dependencies. The defaults set the trap. SQS queues default to a 30-second visibility timeout. Lambda functions routinely get their timeout bumped to 60, 120, or 900 seconds as they grow. Nobody bumps the queue at the same time, because the two settings live in different consoles, different IaC resources, and usually different pull requests. The failure signature points elsewhe

Every developer has that one project. The passion build that sits in the back of your mind for months—or even years—before you finally sit down, crack your knuckles, and make it a reality. For me, that project was building a modern, open-access bilingual digital lexicon bridging English and Assamese: AssameseDictionary.org . While it started as a personal milestone dream, it quickly turned into a massive data engineering and architecture challenge. Here is how I tackled parsing a massive vocabulary database and serving it globally with near-zero latency. 🏗️ The Core Challenge: Scale vs. Speed A dictionary isn't like a standard SaaS app or landing page. It lives and dies by its database depth. To make this a truly definitive tool, I compiled, cleaned, and programmatically validated an extensive vocabulary index mapping over 293,000 words . The dataset doesn't just hold simple translations; it maps complex bidirectional lookups, phonetic transliterations, advanced English definitions, context usage examples, and cross-linked synonym tokens. If I threw this massive dataset into a traditional relational database hooked up to a standard server setup, I ran into immediate roadblocks: Latency: Heavy search queries on a dataset this size can cause noticeable lag. Cost/Overhead: Maintaining and scaling database servers for unpredictable public traffic gets expensive fast. I wanted the search utility to snap back instantly. To achieve that, I had to ditch traditional server paradigms entirely. ⚡ The Architecture: Serverless Edge Caching To keep things ultra-lightweight, highly cost-effective, and blazing fast, I built the platform around an edge-computing topology: The Runtime: I offloaded the backend logic entirely to Cloudflare Workers . Instead of routing traffic to a centralized origin server, queries are intercepted and executed at serverless edge locations physically closest to the user. The Data Layer: Instead of an active SQL database bottleneck, I mapped the data mat

Let's be honest—traditional A/B testing is broken. If you've ever implemented client-side testing, you know the drill. Your users load the page, wait for JavaScript to execute, and then—flicker—the content changes. It's jarring. It hurts your Core Web Vitals. And worst of all, your experiments might be measuring user frustration instead of genuine engagement. But what if you could run A/B tests with zero flicker? What if you could modify your HTML before it even reaches the browser? That's exactly what Cloudflare Workers make possible . The Server-Side Advantage A/B testing at the edge means making decisions about which variant a user sees before any HTML is sent to their browser . Instead of: Load the page Execute JavaScript Flicker Apply the variant Track the result You get: Decision made at the edge Correct HTML streamed immediately Zero flicker Better performance Companies like Ninetailed are already using Cloudflare Workers to achieve 2-3x cost savings compared to traditional serverless platforms, all while delivering personalized experiences with minimal latency . How It Actually Works The concept is surprisingly elegant. Your Cloudflare Worker intercepts incoming requests, checks for a cookie to maintain user consistency, and routes users to the appropriate variant . Here's a simplified version that's production-ready: javascript const NAME = "myExampleWorkersABTest"; export default { async fetch(req) { const url = new URL(req.url); // Determine which group this user is in const cookie = req.headers.get("cookie"); if (cookie && cookie.includes(`${NAME}=control`)) { url.pathname = "/control" + url.pathname; } else if (cookie && cookie.includes(`${NAME}=test`)) { url.pathname = "/test" + url.pathname; } else { // New user—randomly assign them (50/50 split) const group = Math.random() < 0.5 ? "test" : "control"; url.pathname = `/${group}` + url.pathname; // Fetch and modify the response to set a cookie let res = await fetch(url); res = new Response(res.body, res

My "serverless" database was billing me like it never slept. Neon has this great feature called scale-to-zero. Your Postgres compute suspends when it's idle, so you only pay for the queries you actually run. For a pre-revenue product, that should cost a few cents a month. Mine ran 24/7. The compute never once scaled to zero. The culprit wasn't my app logic. It was my database driver. I was using postgres-js, which holds a persistent connection open to the database. From Neon's side, an open connection looks like activity, so it never suspended. It just stayed awake and kept quietly billing me. The fix was basically one conceptual change: switch to @neondatabase/serverless, the HTTP driver. Instead of a long-lived connection, every query becomes a stateless one-shot HTTP request. The query fires, the connection closes, and the compute is free to suspend. Scale-to-zero finally worked. The lesson I keep relearning as a solo founder: "serverless" is a property of your whole stack, not a checkbox on one service. One persistent connection upstream and the whole cost model breaks.

The access you revoked in your last review is probably still live. I know how that sounds, but it is how most access recertification actually works. A tool generates a list, an owner clicks approve or revoke, the cycle gets marked complete, and then nothing touches the real resources. The review produces a record. The permissions stay exactly where they were. You attested to a state that was never made true. That gap bothered me for a long time, so I built something to close it and open-sourced it on AWS's aws-samples org. It is called VIGIL. The core idea A normal review answers one question: should this access still exist? The owner says no, a ticket gets filed, and maybe someone actions it next quarter. Between the decision and the change, the risk just sits there. I wanted the decision and the change to be the same step. So VIGIL does four things: It discovers resources by their owner tag and works out who actually has access to each one. It asks the owner to keep, trim, or remove that access. It applies that decision on the live resource. It records what happened in a way you can later prove. The part I care about most: scoped enforcement The lazy way to revoke someone's access to one bucket is to detach their policies. That nukes their access to everything, and it is how you cause an incident while trying to improve security. VIGIL never does that. If the access came from a bucket policy, it removes just that principal, or just the specific actions, from that bucket's policy. If the access came from the principal's own IAM policy, it adds a resource-scoped explicit Deny instead of touching shared policy, so nothing else the principal can do is affected. In practice it can remove only s3:PutObject for one principal on one bucket and leave everything else alone. If a change cannot be made safely and narrowly, it raises a ticket instead of guessing. I would rather it do nothing than do something broad. Making enforcement durable Enforcement is not a synchronous c

I built a zero-knowledge secret sharing tool. Text and files are encrypted in the browser with AES-GCM 256 before upload - the server only ever sees ciphertext. The decryption key lives exclusively in the URL fragment (#k=...). URL fragments are never sent in HTTP requests (RFC 9110 §4.2.3), so Cloudflare Workers, D1, and R2 never see it - even in logs. A few things I tried to do right: Single-use by default: Durable Objects handle atomic read→decrement→delete with blockConcurrencyWhile, no race conditions on concurrent requests Paranoid mode: returns not_found instead of expired/burned, no timing oracle Revoke endpoint: delete a drop before it's read using a SHA-256'd token with constant-time comparison CLI: burnafter send / burnafter receive - full E2E from terminal, key never touches a browser - /security page with a live in-browser AES-GCM demo and a manual Node.js decryption snippet so you can verify without trusting me Stack: Cloudflare Workers + D1 + R2 + Durable Objects. No third-party crypto libs. Live: https://burnafterread.casablanque.com Source: https://github.com/casablanque-code/burnafterread Verify: https://burnafterread.casablanque.com/security

What just happened On June 22, 2026, AWS quietly launched Lambda MicroVMs. Not a Lambda feature update. A new compute primitive sitting between Lambda Functions (stateless, 15-min max) and EC2 (full VM, you manage everything). Each MicroVM is an isolated Firecracker VM with its own HTTPS endpoint, running your code from a pre-built snapshot. Stateful. Up to 8 hours. Suspend when idle, resume on demand. I tested it the same week. Here's what I found. The test setup A minimal Python HTTP server packaged as a Dockerfile: from http.server import HTTPServer , BaseHTTPRequestHandler import json , time , os class Handler ( BaseHTTPRequestHandler ): start_time = time . time () request_count = 0 def do_GET ( self ): Handler . request_count += 1 body = json . dumps ({ " message " : " Hello from Lambda MicroVM! " , " uptime_seconds " : round ( time . time () - Handler . start_time , 2 ), " requests_served " : Handler . request_count , " pid " : os . getpid () }) self . send_response ( 200 ) self . send_header ( " Content-Type " , " application/json " ) self . end_headers () self . wfile . write ( body . encode ()) HTTPServer (( " 0.0.0.0 " , 8080 ), Handler ). serve_forever () The Dockerfile: FROM public.ecr.aws/lambda/microvms:al2023-minimal RUN dnf install -y python3 && dnf clean all WORKDIR /app COPY app.py . EXPOSE 8080 CMD ["python3", "app.py"] How it works Three steps: Zip code + Dockerfile → upload to S3 create-microvm-image builds the container, starts the app, takes a Firecracker snapshot of memory and disk run-microvm launches from that snapshot Every launch resumes from the pre-initialized state. No cold boot. Your app is already running the moment the MicroVM starts. aws lambda-microvms create-microvm-image \ --name hello-microvm-test \ --code-artifact "uri=s3://my-bucket/artifact.zip" \ --base-image-arn arn:aws:lambda:us-east-1:aws:microvm-image:al2023-1 \ --build-role-arn arn:aws:iam::123456789:role/MicroVMBuildRole Image build took about 3 minutes. Once done: aw

TL;DR for vibe coders: Shipped an app with Cursor, Claude Code, or v0 and got a scary Vercel or Neon bill? You probably don't need a bigger plan. You need a few fixes. Not technical? Copy the agent-rules folder from the companion repo into your project and tell your AI editor "apply these cost rules." That's the whole job. You don't need Vercel Pro. You don't need to "Launch" on Neon. A lot of the time you need the opposite of an upgrade. You need your stack to go to sleep. Here's the moment that started this. A Neon bill landed, across three small Next.js apps running on Vercel with a Neon Postgres database, and the breakdown was lopsided in a way that gave the whole game away. $32.65 of it was compute. Storage was 5 cents. History and data transfer were zero. The meter said 308 hours of compute time in about three weeks, which is a database awake for roughly fifteen hours a day, every day. The instinct when a bill climbs is "I must be getting traffic, time for a bigger plan." So before doing that, I opened Google Analytics. Zero users in the last 30 minutes. Meanwhile the database compute was pinned active, and had been more or less around the clock. So this was never a storage problem or a traffic problem. Almost the entire bill was one thing: paying for a database to stay awake doing nothing. Both of those things were true at the same time, and here's the part I'll say up front so nobody has to guess: I didn't hand-write the mistakes that caused this. AI coding agents did. I described features, the agent shipped working code, and that code carried a handful of patterns that quietly defeat scale-to-zero. That's not a confession of bad engineering. It's just what building looks like in 2026. Most of us are vibe-coding onto serverless and cloud infra now, and the agent optimizes for "make the feature work," not "keep the database asleep when no one's around." It has no idea your compute bills by the hour it stays awake, so it has no reason to care. That's the whole

AWS Hero Vadym Kazulkin explains how to overcome Java’s enterprise hurdle on AWS Lambda: cold starts and memory footprints. He shares a technical deep dive into performance tuning, comparing fully managed AWS SnapStart (with pre-snapshot priming hooks) against GraalVM ahead-of-time compilation, while addressing the latest architectural implications of Project Leyden and Java 25. By Vadym Kazulkin

Modern enterprise order processing architectures must decouple synchronous client demands from asynchronous backend dependencies. Here I'll detail a highly scalable, fault-tolerant design built on AWS. By utilizing an automated API Gateway entry point, specialized Amazon Cognito authentication, optimized AWS Lambda logic blocks, an engineered RDS Proxy connection layer, and an event-driven SQS/EventBridge core, the design guarantees isolation, cost efficiency, and sub-millisecond structural routing. The Scenario User places an order → payment is processed → inventory updated → confirmation email sent Client → API Gateway (Cognito auth + validation) → Order Lambda (business logic + DynamoDB write) → SQS (payment queue) → Payment Lambda → EventBridge → Inventory Lambda → Notification Lambda (SES) 1- Entry Point — API Gateway REST endpoint: POST /orders Request validation via API Gateway models (reject malformed payloads instantly, no Lambda invoked) Auth via Cognito User Pool Authorizer — validates JWT token on every request How It Works The Request Hits: A client sends a POST /orders request with a JWT token in the header. Auth Check: API Gateway automatically intercepts the request and validates the JWT against the Cognito User Pool. If expired or spoofed, it returns 410 Gone / 401 Unauthorized right there. Payload Check: Next, it compares the body against your JSON Schema Model. If a required field like customer_id is missing, API Gateway instantly drops it with a 400 Bad Request. The Win: Your downstream services (like Lambda) are never invoked for bad/unauthorized requests, saving compute costs and protecting against basic DDoS or bad actor spam. Gotchas Cognito Latency: While Cognito authorizers are native, they can add a slight latency overhead to your API's P99 metrics during peak traffic. For massive global scale, some enterprises migrate to custom Lambda Authorizers that cache tokens in ElastiCache (Redis). Model Validation Limits: API Gateway's built-in val

When I built whatsmy.fyi , I assumed I'd need a geolocation provider: MaxMind, ipinfo, ip-api, pick your poison. They all mean the same thing: an external dependency, an API key, a quota, added latency, and someone else's server seeing your users' IPs. Then I found out Cloudflare Workers makes the whole category unnecessary. The cf object Every request that hits a Cloudflare Worker carries a request.cf object, populated at the edge before your code even runs. No lookup, no latency, no key. Here's what's inside: { asn : 34984 , // ISP's autonomous system number asOrganization : " Superonline " , // ISP name city : " Istanbul " , region : " Istanbul " , country : " TR " , continent : " AS " , isEUCountry : undefined , // "1" if EU, undefined otherwise latitude : " 41.01380 " , // string, not number! longitude : " 28.94970 " , postalCode : " 34000 " , timezone : " Europe/Istanbul " , colo : " IST " , // which CF datacenter handled this clientTcpRtt : 12 , // user's RTT to the edge, in ms httpProtocol : " HTTP/3 " , tlsVersion : " TLSv1.3 " , tlsCipher : " AEAD-AES128-GCM-SHA256 " } That last group surprised me most: you get the user's HTTP protocol, TLS version, and actual TCP round-trip time for free. Try getting that from a geo API. A complete IP endpoint in ~30 lines export default { async fetch ( request ) { const cf = request . cf ?? {}; const ip = request . headers . get ( " CF-Connecting-IP " ); return Response . json ({ ip , city : cf . city ?? null , country : cf . country ?? null , isp : cf . asOrganization ?? null , asn : cf . asn ?? null , timezone : cf . timezone ?? null , lat : cf . latitude ? parseFloat ( cf . latitude ) : null , lng : cf . longitude ? parseFloat ( cf . longitude ) : null , protocol : cf . httpProtocol ?? null , tls : cf . tlsVersion ?? null , rttMs : cf . clientTcpRtt ?? null , }); }, }; That's the entire backend. No database, no GeoIP file to update monthly, no vendor. The gotchas (learned the hard way) 1. Coordinates are strings. lati

Redis is fast. But self-hosting Redis on a serverless stack is a nightmare — cold starts, connection pool exhaustion, and managing a persistent server that your serverless functions keep hammering. Upstash solves this with an HTTP-based Redis API that scales to zero, charges per request, and works natively with Next.js App Router. This guide covers the patterns that actually matter in production: cache-aside with proper TTLs, SWR (stale-while-revalidate), session storage, and pub/sub. Real code, real trade-offs. Read the full article with all code examples at stacknotice.com Why Upstash Over a Traditional Redis Instance Standard Redis uses persistent TCP connections. Serverless functions don't maintain persistent connections — every invocation potentially opens a new one. At scale, you hit ECONNREFUSED or max connection errors that are annoying to debug and expensive to fix. Upstash's @upstash/redis client talks over HTTP/REST. No connection pool, no connection limit headaches. Each request is stateless. This is exactly what Next.js Server Components and Route Handlers need. Other advantages: Pay per request — a cache that never gets hit costs $0 Global replication — low latency from any Vercel edge region Native Edge Runtime support — works in Next.js middleware Free tier — 10,000 commands/day, no credit card needed Setup npm install @upstash/redis // lib/redis.ts import { Redis } from ' @upstash/redis ' export const redis = new Redis ({ url : process . env . UPSTASH_REDIS_REST_URL ! , token : process . env . UPSTASH_REDIS_REST_TOKEN ! , }) Pattern 1: Cache-Aside // lib/cache.ts import { redis } from ' ./redis ' export async function withCache < T > ( key : string , fetcher : () => Promise < T > , options : { ttl ?: number ; prefix ?: string } = {} ): Promise < T > { const { ttl = 300 , prefix = ' cache ' } = options const cacheKey = ` ${ prefix } : ${ key } ` const cached = await redis . get < T > ( cacheKey ) if ( cached !== null ) return cached const data = awai

Why Your Vector Database Is Overpriced: Lucene's 32x Compression and Serverless Economics In 2026, the boundary between "search engine" and "AI infrastructure" has dissolved. What started as text indexing has become the backbone of retrieval-augmented generation, vector databases, and serverless AI pipelines. This is the story of how the oldest search technology in the Java ecosystem became the most important infrastructure you've never noticed. The Convergence No One Saw Coming Five years ago, if you said Apache Lucene would power the next generation of AI infrastructure, you'd have been laughed out of the room. Lucene was the boring Java library that powered Elasticsearch — reliable, yes, but hardly exciting. The action was in vector databases: Pinecone, Weaviate, Qdrant. The cool kids had moved on. That narrative died in 2025. What happened was a structural inversion. While vector-native databases optimized for one thing (fast similarity search), the real production pain points were everywhere else: hybrid search, metadata filtering, provenance tracking, multi-tenant security, and — most critically — the ability to query both your documents and your vectors in a single, unified system. Lucene didn't just survive this transition. It engineered it. Through a series of aggressive, hardware-native optimizations between versions 10.0 and 10.4, Lucene transformed from a text indexer into a vector search kernel capable of outperforming specialized databases while maintaining the operational maturity that enterprises actually need. And Elasticsearch, riding on Lucene's coattails, didn't just integrate vectors — it re-architected itself into a stateless, serverless platform that happens to do search. This post examines three layers of that transformation: the engine (Lucene), the platform (Elasticsearch), and the architecture (AI-native search infrastructure). Each layer tells a different story, but they share a common thread: the future of AI infrastructure is being buil

TL;DR If you're shipping AI inference and tired of babysitting GPUs, serverless is the way out. You deploy the model, the platform scales it from zero to hundreds of GPUs and back, and you only pay for the time you actually use. If I'm picking one to start with, it's DigitalOcean . It's got the widest GPU lineup of any serverless provider (RTX 4000 Ada all the way up to NVIDIA Blackwell B300 and AMD's MI350X), one API and one bill instead of five, and it's simple enough to ship on without a sales call. (More on why that one's personal for me below.) Below I compare 9 providers across the things that actually matter: GPU specs, per-hour pricing, cold-start latency, model support, and how nice they are to build on. DigitalOcean, RunPod, Modal, Koyeb, Together AI, Replicate, Baseten, Fal, and Cloudflare Workers AI each win at something different, from cheap experimentation to global edge inference. Contents Why I ran this The field at a glance How I evaluated these providers Per-provider analysis: DigitalOcean RunPod Modal Koyeb Together AI Replicate Baseten Fal Cloudflare Workers AI Why I keep coming back to DigitalOcean The short version Questions I actually get asked Why I ran this Quick note on why this exists. At work I get a front-row seat to a lot of people shipping an AI model into production for the first time: students, first-time founders, my own team. And lately the same question keeps coming up: where do I actually run this thing? I was tired of answering with a shrug and "it depends," so I did the homework myself. Signed up, read the pricing pages, ran the comparisons, and wrote it all down. Nobody's a real expert at this yet, me included, so I'd rather share my notes and get corrected than pretend I've got it figured out. And here's the thing about AI inference in 2026: demand blew past what the old way of provisioning GPUs can handle. Teams that used to wait weeks for dedicated hardware now need a model live in minutes. The ground moved. And the stuff t

Let me tell you exactly what happened the first time I tried to set up Lambda manually. Four hours. IAM trust policies I didn't fully understand, ARNs copy-pasted into the wrong fields, an API Gateway that was technically configured but somehow not routing anything correctly, and a deploy that failed with an error message pointing me nowhere useful. I hadn't written a single line of actual business logic yet. That's when someone on my team mentioned the Serverless Framework. My first reaction was honestly skepticism — another abstraction layer sounded like another thing to learn and eventually fight with. I was wrong about that. This isn't a "look how clean this tool is" post. It's more like: here's what I actually did to get a Postgres-backed CRUD API running on Lambda, step by step, including the parts that tripped me up. What the Framework Is Actually Doing Under the Hood Worth knowing before you start: the Serverless Framework isn't magic. It's generating CloudFormation templates and submitting them to AWS on your behalf. Your Lambda functions, API Gateway routes, CloudWatch log groups — all of it gets provisioned from a single config file. It works with other providers too, but the AWS integration is where it really earns its keep. The console clicking and manual ARN-wiring that burns time at the start of every serverless project? Gone. Same deploy workflow whether you're building a REST API, an event processor, or a cron job. Once you've done it once, the second project takes a fraction of the time. What You're Building Four live endpoints backed by PostgreSQL. A Users table. Create, read, update, delete — nothing exotic, but a real enough foundation that you can extend it into something actual once this guide is done. You'll need an AWS account, the AWS CLI installed, and the Serverless Framework installed before starting. That's it. Step 1: Sort Out Your AWS Credentials Run this to create both config files in one go: bash cat << EOF > ~/.aws/credentials [def