Start a project →
AI · Machine learning · Data · IT

Clear paths through complex data

Amber Maze designs, builds and runs AI, machine-learning and data systems for businesses. We take a tangled problem, map it, and ship the shortest route to something that works in production.

Talk to us → See our products
STATE carvingCELLS 0PATH —
Machine learningLLM agentsAI harness engineeringEU AI Act readinessData engineeringForecastingProcess automationAnalytics & BIMLOpsComputer-assisted valuation Machine learningLLM agentsAI harness engineeringEU AI Act readinessData engineeringForecastingProcess automationAnalytics & BIMLOpsComputer-assisted valuation
Services

Everything data, from the first query to the last mile.

We work where data meets the business: the pipelines that feed it, the models that learn from it, the agents that act on it, and the rules that govern it. One team, end to end.

01

AI & LLM solutions

Assistants, agents and retrieval systems built on your own documents and data, wired into the tools your team already uses.

  • Agents
  • RAG
  • Document AI
  • Copilots
02

Machine learning

Forecasting, pricing, valuation, classification and anomaly detection models that are measured, monitored and retrained.

  • Forecasting
  • Valuation
  • Anomaly detection
  • MLOps
03

Data engineering

Pipelines, warehouses and data models that turn scattered spreadsheets, APIs and databases into one source you can trust.

  • ETL / ELT
  • Warehousing
  • Data quality
  • APIs
04

EU AI Act readiness

We classify your AI systems by risk, close the gaps, and leave you with the documentation, oversight and logging the regulation asks for.

  • Risk classification
  • Tech docs
  • Governance
  • AI literacy
05

AI harness engineering

The layer around the model that makes it dependable: tools, memory, guardrails, evaluation suites and monitoring, so an agent does the job the same way every time.

  • Agent harnesses
  • Evals
  • Guardrails
  • Observability
06

IT solutions & automation

Internal tools, integrations and automated workflows that remove the manual steps between systems, plus dashboards people actually open.

  • Automation
  • Integrations
  • Dashboards
  • Web apps
EU AI Act

Compliance built in, not bolted on.

The EU AI Act sorts AI systems by the risk they carry and sets obligations for each tier. We map every system you run or plan to build, tell you where it falls, and build the controls into the system itself.

  • AI system inventory
  • Risk classification
  • Technical documentation
  • Human oversight design
  • Logging & traceability
  • Transparency notices
  • Data governance
  • AI literacy training
UnacceptableProhibited practices. We check that nothing you run falls here.
High riskStrict requirements: risk management, data quality, documentation, oversight.
Limited riskTransparency duties, such as telling people they are talking to an AI.
Minimal riskMost systems. Voluntary codes of conduct and good practice.
General-purpose AISeparate obligations for the models themselves, and what that means when you build on them.
Products

We build our own products, too.

Running products in production keeps our consulting honest. The same team, stack and standards go into client work.

Real estate · AI valuation

VERICITY

Find out any property's value, history and surroundings.

A free real-estate platform that combines AI property valuation with official transaction records, cadastre data, and flood and noise maps. In Latvia it covers the whole country using State Land Service (VZD) data, with no registration needed, and the same platform now runs in Slovenia, Poland and Spain.

11,339Riga transactions in the last 12 months
1,144 €Riga average per m², last 12 months
4Markets: Latvia, Slovenia, Poland, Spain
Available in
EXAMPLE
Estimated value0 €
—
TransactionsCadastreFlood mapNoise map
Customer insight · Live feedback

Inquiso

Turn live customer interactions into insights.

Inquiso captures survey and feedback data from in-person and field interactions and analyses it as it arrives, so teams understand customer sentiment the moment it changes, not weeks later in a report.

LiveCapture and analysis in real time
FieldBuilt for in-person interactions
AISentiment read automatically
Open inquiso.app →
EXAMPLE
Sentiment, live +0.00
POS
NEU
NEG
Responses 0
Approach

Five steps from maze to route.

Every project follows the same path. Each step ends with something you can see, test or use, so there are no long silent stretches.

1

Map

We learn the business question, the data you have and the decision it should change.

2

Probe

A quick prototype on real data shows what is possible before anyone commits a budget.

3

Build

Pipelines, models and interfaces, built in short cycles you can review.

4

Prove

Evaluation against agreed metrics, plus the documentation the AI Act may require.

5

Run

Deployment, monitoring and handover, or we keep running it for you.

About

A small, senior team that ships.

Amber Maze is a Latvian company founded by a data science lead with a background in quantitative finance and production forecasting systems. We work in Latvian and English, with clients across the EU.

We stay small on purpose: the people you talk to are the people who write the code.

01

Production or it didn't happen

We measure success by systems people use daily, not by slide decks.

02

Numbers before opinions

Every model ships with an evaluation you can read and repeat.

03

Your data stays yours

EU hosting by default, clear access rules, and no training on client data without consent.

04

Plain language

No hype. We tell you what the model can and cannot do.

Got a maze of data? Let's find the path.

Tell us what you are trying to decide, automate or predict. We reply within two working days with a first view and next steps.