Web platformReal estate · CRE
Relm: multi-agent AI that finds hidden deal risk in commercial real estate
Minutesaddress to Intel Report
Cloud & Backend Architecture
Infrastructure that scales — and costs less.
Scalable backends, cloud architecture, DevOps and cost optimization on AWS, Azure and GCP, so your product stays fast, secure and affordable as it grows.
Audit
cost, risk and bottlenecks first
IaC
everything in code
Alerts
wired in before launch
What you getEverything to go from idea to production.
01
Architecture design
Services, data and event flows designed for scale.
02
Cost optimization
Right-sizing, reserved capacity and waste removal.
03
DevOps & CI/CD
Automated builds, tests and zero-downtime deploys.
04
Infrastructure as code
Terraform or Pulumi for repeatable environments.
05
Security & compliance
IAM, encryption, backups and SOC 2 readiness.
06
Observability
Logs, metrics, tracing and on-call alerting.
How we deliver
- STEP 01
Audit
Current cost, risk and bottlenecks.
- STEP 02
Design
Target architecture and plan.
- STEP 03
Migrate
Phased, zero-downtime moves.
- STEP 04
Operate
Monitor, optimize, support.
LLM costsKeeping LLM inference costs down.
On an AI product, model calls can outgrow the rest of the cloud bill. Four levers cut them without changing what users see; three are published price multipliers, from Anthropic’s pricing page, checked October 18, 2026.
Lever 1
Batch background jobs
Work that can wait, such as nightly summaries or bulk classification, goes through the Batch API at 50% off both input and output tokens.
Lever 2
Cache repeated context
A cache hit costs 0.1x the base input price on most Claude models, 0.05x on Sonnet 5.5 and Opus 5.5, and 0.025x on Fable 5.1 and Mythos 5.1. A 5-minute cache write costs 1.25x, so a long system prompt pays off after one reuse.
Lever 3
Route by difficulty
A small model handles routine requests and hands hard ones to a larger model. The routing rule is tested on the same evaluation set as the product.
Lever 4
Price in residency
US-only inference costs 1.1x the standard rate (10% more) on every token category on Claude 4.6 and later models. Turn it on when a contract requires it, not by default.
Self-hosting
A model on your own GPUs swaps per-token fees for a fixed bill. It pays off at steady, high volume, or when policy rules out external model calls. Private RAG deployment
Where the numbers come from
Anthropic’s published API prices. The multipliers stack with each other, and rates change, so we re-check them in every audit. Anthropic pricing
Our rule: give every AI feature a monthly cost budget with an alert before reaching for a cheaper model.
Build or buyManaged platform or your own cloud architecture?
Managed platforms are the right start. Own the architecture once cost or limits start to hurt.
| Your situation | Our recommendation |
|---|---|
| An early product with a small team | Stay on a managed platform like Vercel or Firebase |
| Growing traffic across several services | Containers on your cloud, defined in code |
| Spiky or event-driven workloads | Serverless functions and queues |
| A cloud bill growing faster than usage | Cost audit first |
Our recommendation: stay managed until the bill or the limits hurt, then move to infrastructure you own, in code.
Typical engagementWhat cloud work costs and how long it takes.
- 01 · 2 weeks
Discovery Sprint
$5,000fixed fee · 2 weeksAn audit of current cost, risk and bottlenecks, with a target architecture and migration plan.
- 02 · 6–16 weeks
Fixed-Price Build
$25k–$120kper project · 6–16 weeksPhased migration or new backend, all infrastructure as code, with zero-downtime deploys.
- 03 · Ongoing
Dedicated Team
From $12kper month · cancel anytimeMonitoring, cost reviews and on-call support with agreed response times.
Built by one in-house team, from discovery to launch and support. AI makes us fast; a senior engineer decides what goes live. Meet the team
What moves the price: the number of services, environments and data stores, and how much downtime the migration can tolerate. Compare engagement models
Selected workShipped with this.
Web platform
iOS · Android · WebMedia · Sports streaming
DubCam: live streaming where fans fund the teams they love
50%revenue share to creators
GuidesRead before you build.
- AI strategy · 7 min readWhy most AI pilots stall before production, and how to ship yours
- Buying software · 11 min readRAG development cost in 2026: what drives the price
- Voice agents · 10 min readWhy sub-second latency makes or breaks a voice agent
- Buying software · 12 min readAI agent development cost in 2026: real price ranges
FAQCommon questions.
Can you reduce our cloud bill?
The audit shows where the money goes before we promise a number. We check right-sizing, reserved or committed-use capacity, idle and orphaned resources, storage tiers and data egress, then price each fix against its saving in your own billing data.
Do you migrate between clouds?
Yes. We move one service at a time, run the old and new versions side by side, shift traffic gradually and keep a rollback path at each step, so users never see a hard cutover.
Do you provide on-call support?
Yes, through a monthly support retainer. Response times, hours of cover and escalation contacts are agreed in writing, and alerts go to your team and ours.
How much does cloud architecture work cost?
Most engagements start with a 2-week Discovery Sprint ($5,000 fixed fee, credited to your build), followed by a fixed-price build ($25k–$120k per project, 6–16 weeks) or a dedicated team (from $12k per month). The build price is fixed after discovery and billed in milestones tied to demoed features. For cloud work, the Discovery Sprint is the audit and target architecture. Your cloud provider bills you directly, in your own account, for the infrastructure itself.
How long does a cloud migration take?
The audit and target architecture take two weeks. A phased migration is a fixed-price build of 6–16 weeks, moving one service at a time with a rollback path at each step.
Which cloud should we choose: AWS, Azure or GCP?
Usually the one your team, customers and credits already favor. Azure suits Microsoft-heavy companies and Azure OpenAI; AWS has the widest service range and Bedrock; GCP is strong for data and Vertex AI. We keep core services portable.
How do you control LLM inference costs?
By routing simple requests to smaller models, caching repeated prompts and embeddings, batching background jobs, setting per-feature budgets with alerts, and self-hosting models only where volume justifies the GPU cost.
Who owns the infrastructure and the access?
You do. Everything runs in your cloud accounts and is defined in Terraform or Pulumi in your repository, so environments are reproducible and nothing depends on our access.
Do you help with SOC 2 readiness?
Yes, on the technical side: IAM, encryption, backups, logging and change management set up the way auditors expect. The audit itself is done by an independent firm; we are not an auditor.
Can you keep our data and models in an EU region?
Yes. We deploy into an EU region such as Frankfurt or Ireland and keep storage, backups and logs there. For model calls we pick your provider’s EU processing option, for example Google Vertex AI in its EU multi-region for the models that support it, or Azure OpenAI in the EU Data Zone, and pin the region in your infrastructure code so it can’t drift.
