Amazonia Beats · Platform engineering
Platform engineering for operations that leave no room for error.
Amazonia Beats builds its own technology and the platforms that stand on it. At the centre is METIS ENGINE, an editorial intelligence framework: it decides what to publish, checks whether it holds up against the source, and flags what is already live when it stops being true. Around it, multi-tenant SaaS platforms for accounting, taxation and public procurement — with every engineering decision recorded and auditable.
- Java 21
- Spring Boot
- PostgreSQL
- Google Cloud
- Terraform
- OpenTelemetry
Live panel
The market we work in, updating itself.
This panel is built with AB Media Stream Dashboards, our own product. Figures come from declared public sources; headlines are collected hourly by the site itself — PHP, MySQL and the host scheduler, with no cloud service in between.
The foundations of our delivery
Not commercial promises. Architecture decisions on the record, with the context behind them and the alternatives that were rejected.
Per-customer isolation
Every customer runs on a dedicated database. Isolation is an architectural boundary, verified by automated tests.
Declarative infrastructure
All infrastructure is described in Terraform and version-controlled. Reproducible environments and fully auditable changes.
Continuous delivery
Continuous integration with an automated suite as a merge condition. Production releases without downtime windows.
Cost efficiency
Managed services with elastic scaling and no idle resources. Budget predictability for the contracting party.
Areas of practice
Three fronts, one foundation: METIS ENGINE, the editorial intelligence framework we built, which the in-house products stand on.
Multi-tenant SaaS platforms
Systems licensed per module on a shared base of identity, billing, permissions and audit. Customers activate what they use and expand scope without replacing the platform.
Integration with government systems
Electronic invoicing, tax regimes, municipal service tax and the federal procurement portal. Fiscal and procurement rules are implemented inside the product, on a versioned rules engine.
Editorial intelligence as in-house technology
METIS ENGINE decides what to publish, checks whether it holds up against the source, and flags for review what is already live when the norm changes. It is the base of AB Media Stream Dashboards, the ebook generator and, in unification, MAV TV Video Maestro.
Architecture
Modern architecture, from the first commit to day-two operations.
Our platforms are cloud-native by design: managed services, elastic scaling, asynchronous messaging and observability from the first deploy. The diagram below is the standard applied across every product we build.
- 01
Edge
Global CDN, managed TLS and routing by customer subdomain.
- 02
Application
Containerised services with elastic scaling and on-demand provisioning.
- 03
Platform
Identity, tenancy, billing, permissions and audit shared by every module.
- 04
Data
A dedicated PostgreSQL database per customer, encrypted at rest, with managed backups and tested restores.
- 05
Integration
Asynchronous messaging, scheduling and connectors to public administration systems.
- 06
Operations
Declarative infrastructure, continuous delivery and standardised end-to-end telemetry.
Artificial intelligence and in-house technology
We do not resell someone else’s AI. We built the layer that governs it.
The market sells text generation. For anyone working with regulation, generating is not the problem — knowing when what was already published stopped being true is. That is what METIS ENGINE is for: the model writes, the engine decides. Validity checking and approval are deterministic code, and the AI never judges its own claim.
Language models in production
Integration with frontier models, version-controlled prompt orchestration and continuous evaluation of answer quality.
Retrieval augmented by validity
Semantic search over our own corpus. Each retrieved passage carries the norm that supports it and the date until which it holds — anything expired never reaches the context handed to the model. In production today for agenda deduplication; under construction for legislation and tender documents.
Structured document extraction
Reading invoices, tender notices and scanned documents, turned into verifiable data ready for reconciliation.
Agents under human supervision
Automation of repetitive tasks with a mandatory approval step and an audit record of every action taken.
Evaluation and guardrails
Regression tests for prompts, explicit scope limits and hallucination checks before anything is published.
Privacy by design
Client data never trains a third-party model. Processing in a defined region and secrets held in a managed vault.
Special projects
A dedicated practice for work outside the catalogue: integration with public-sector systems, legacy data migration, automation of critical processes and short-deadline proofs of concept. Senior staff assigned, formalised scope and verifiable incremental delivery.
- Public-sector system integration
- Legacy data migration
- Critical process automation
- Short-deadline proof of concept
Proprietary platform
IMPERIO
A SaaS platform for accounting firms and in-house tax departments, licensed per module. Under development, with the first two modules being built.
Tax
Tax assessment and ancillary obligations, management of tax liabilities, and settlement programmes with interest projection.
Procurement
Tender monitoring, screening by fit with the company profile, and deadline control throughout the bidding process.
Engineering principles
Rules that apply to every project in the company, established before they were needed.
- 01
Every structural decision is recorded
Each significant choice produces an ADR with context, rejected alternatives and consequences. Decision traceability is part of the delivery.
- 02
Secrets never reside in the repository
No credential, certificate or password is version-controlled, not even in test environments. Secrets live in a managed vault with access on the record.
- 03
Customer data is a third-party asset
Invoices, digital certificates and accounting records belong to the client of the contracting party. They are encrypted at rest and never written to logs.
- 04
Mandatory review, small increments
Changes that are small, reviewable and covered by tests. Nothing is released without a passing continuous integration run.
Does your organisation need a supplier that documents every decision?
We serve private and public contracts. Send us the scope, the tender notice or the technical requirement: we respond with an objective assessment of feasibility, timeline and limits.