WHEN IT FITS
Four signals
no pack will fit.
Qualification criteria, not a pitch. If one of these lines describes your case, Custom is the right door.
Integration with 10+ systems
Banking core, ERPs from different generations, mainframe, internal platforms with their own rules. Each connection is design work, not plug-and-play.
Vertical platform
An internal or market product in a specific sector such as health, energy, or legal, where the AI has to understand the domain ontology, its rules, and its flow.
Model under regulatory constraint
BACEN, ANS, ANEEL, LGPD in a sensitive context. Auditing, governance, White Box AI, and a full paper trail are entry requirements.
AI thesis with no ready product
A strategic hypothesis that defines your competitive advantage, which is exactly why no one has packaged it yet. The solution becomes your intellectual property.
HOW WE APPROACH
Consultative on the outside.
Engineering on the inside.
No open-ended "let's see," no pre-fab waterfall. It is a lean process, with a structured Discovery and a senior squad plugged in from start to finish.
Structured Deep Discovery
A 2-4 week immersion in the business, mapping of data and systems, prioritized hypotheses, and a baseline metric. Output: a release plan with scope, schedule, and an architecture already validated in the real infrastructure.
Senior multidisciplinary squad
Data engineering, ML, generative AI, and product, all senior and all plugged in. Whoever designs the solution talks to the client. There is no "account head" handing work down to juniors.
Client AWS account from day one
The solution runs inside your infrastructure from the pilot on. Environment isolation, auditable logs, cost control at the source. Nothing sits "in BlueMetrics staging" to migrate later.
Code delivered, no lock-in
Repository, models, pipelines, runbooks, and documentation stay in your account. We leave with no exit fee, no artifact retention, and no license tie-in.
TYPICAL CASES
Four kinds of problem
that land here.
Not an exhaustive list. It is the pattern that keeps recurring across the projects that come in under this model.
Vertical AI platform
An internal or market product specific to one sector, where the AI depends on ontology, rules, and a domain flow that a generic pack does not cover.
Integration with 10+ legacy systems
Core, ERP, CRM, and prior-generation systems, often with no modern API. The work is as much data architecture as it is AI.
Model under a specific regulatory constraint
Sectors with continuous auditing (BACEN, ANS, ANEEL). Governance and explainability are part of the product, not a compliance attachment.
AI thesis with no ready product
The company has found an AI advantage that no one has packaged yet. Custom is how you turn it into IP and get there first.
METHODOLOGY
Four phases.
Measurable value at every stop.
A phase diagram, not a fixed schedule. Each phase has a clear scope and an explicit decision to continue, adjust, or pivot.
Deep Discovery
2-4 weeks
Business immersion, mapping of data and systems, prioritized hypotheses, a baseline metric. Architecture validated in the client's real infrastructure. Output: a release plan.
Pilot / MVP
8-16 weeks
Model trained on real data, minimal integrations, an isolated environment inside the client's AWS account. A measurable result against the baseline before any expansion.
Production
Variable scope
Hardening, integration with core systems, MLOps in production, a rollout plan, and training for the internal team. Supervised operation in the first weeks after go-live.
Continuous optimization
Recurring
Scope expansion, new cases on the same base, retraining, drift and cost monitoring. It can become AI as a Service or stay with the internal team.
DIFFERENTIATORS
What you actually
take home.
The same principles from our sales deck, now tied to what makes this specific model different.
White Box AI
Explainable decisions, a full paper trail, and quality gates from the first sprint, not bolted on later as a compliance attachment.
Optional continuous ops
You decide who runs it after go-live: your own team, AaaS, or another vendor. Nothing locks you in.
Outcome over deliverable
Each phase closes with a metric measured against the baseline. We do not run a POC that has no path to production.
Full code ownership
Repository, models, pipelines, and documentation stay in your AWS account, with no license, no retention, and no exit fee.
Deep AWS expertise
Advanced Partner. SageMaker, Bedrock, Textract, and AWS-first infrastructure in production for over a decade.
RELATED CASES
Outcome, not slide.
Three projects delivered as Custom. Real metrics, client account, code delivered.
STACK & GOVERNANCE
AWS-first.
Governance as a product.
We are an AWS Advanced Partner and build AWS-first when the environment allows. Every solution runs inside the client's AWS account, with environment isolation, auditable logs, and White Box AI from day one. Where it applies, we follow the sector's regulatory frameworks (LGPD, BACEN, ANS, ANEEL). When a client already has a consolidated stack, whether Azure, GCP, on-prem, or hybrid, we inherit what is already running. We do not treat AWS as universal. We treat it as a well-grounded technical default.
QUESTIONS WHILE DECIDING
FAQ
When the problem is singular: integration with 10+ legacy systems, a proprietary vertical platform, a model under a specific regulatory constraint, or an AI thesis with no ready product on the market. If the case matches a recurring pattern, such as service at scale, contract analysis, credit scoring, or forecasting, a Solution Pack is usually the faster path.
Deep Discovery in 2-4 weeks, Pilot/MVP in 8-16 weeks. The first delivery is designed to produce a measurable result in the first cycle, against the baseline set during Discovery. Cases with critical core integration can extend the production timeline, but the validation cadence per phase stays the same.
The client does. The whole solution runs in the client's AWS account from the pilot on, and the full source code is delivered at the end of the project: repository, models, pipelines, runbooks, and documentation. There is no exit fee, no artifact retention, and no license tie-in.
A senior multidisciplinary squad (data engineering, ML, generative AI, product), allocated for the project cycle. There is no "account head" handing work down to juniors; whoever designs the solution talks to the client. The composition shifts by phase, but the technical core stays stable from Discovery to go-live.
The client decides. You can run it with your internal team after a knowledge transfer (pairing, runbooks, handover sessions), contract AI as a Service for continuous operation with a BlueMetrics squad, or move to another vendor. No lock-in, no retention.
No. Deep Discovery already validates the architecture in the client's real infrastructure, and the Pilot/MVP goes into production, even if limited. A POC with no path to production is not our model: it produces a demo, not operational learning.
SIDE EXIT
Does your problem match a pattern?
Service at scale, document analysis, credit scoring, forecasting, omnichannel integration: these are recurring patterns. For those, Solution Packs deliver a pilot in weeks, with customization on top of a validated base.