braingency
EN – Switch to English LV – Switch to Latvian
ML Lab

Braingency ML Lab

Applied machine learning for specific business problems

01 What the ML Lab is

A division, not a side project

The ML Lab is Braingency's dedicated machine learning division. It exists because the models that create real business value are rarely general-purpose — they are trained on a specific company's data to solve a specific decision problem. The Lab builds those models, integrates them into production systems, and conducts our own research in machine learning and AI.

02 What we do
01

Models inside our own products

Our financial ERP platform includes a scoring module where we train machine learning models on real user data from the lender's own portfolio. The resulting model scores incoming credit applications directly within the origination workflow — the model is not a separate analytics product, it is part of how the loan gets approved.

Verteo, our real estate valuation platform, runs on a model built entirely in-house and trained on Latvian property market data. It produces price forecasts used for mortgage collateral assessment and credit scoring inputs.

02

Custom models for clients

We build machine learning models on commission for companies across different industries. The typical engagement starts with the client's existing operational data and a decision they want to make better or faster — risk assessment, demand forecasting, classification, pricing, anomaly detection. We handle data preparation, model development, validation and production integration.

Because our engineering team builds the systems the models run inside, the handover problem that kills most ML projects does not arise: there is no gap between the team that trains the model and the team that has to deploy it.

03

Research

The Lab conducts ongoing research in machine learning and AI, focused on the domains we work in — credit risk modelling, valuation, and applied AI in regulated environments.

The ML Lab operates within the same engineering organisation that builds our production systems. Machine learning capability is embedded across the founding team.

03 Verteo, nationwide

Every cell is a valuation

Verteo estimates a price for every part of the country and re-estimates as sales come in. Where comparable sales are scarce, the range widens — and the model says so.

An illustrative map of Latvia divided into hexagons, shaded by estimated price per square metre: brightest around Riga and Jūrmala, lower in regional towns and lowest in rural areas. The table beside it lists the values for the main towns.

verteo / latvia · €/m²
Illustrative field, not market data
€/m²
    Main towns
    Main towns
    Town€/m²90% range
    Incoming sales
    fig. 09 — estimated price per m², by area
    04 Case examples

    Two models in production

    Case 01

    Credit scoring

    A lending client's underwriting process relied on static rule-based criteria. We trained a scoring model on their historical application and repayment data and integrated it into the origination workflow, allowing applications to be assessed against patterns in their actual portfolio performance.

    Model score by outcome
    • Repaid
    • Defaulted
    Approved
    —
    Default rate
    —
    Good declined
    —
    fig. 10 — the approval trade-off, on synthetic data
    Case 02

    Property valuation

    Verteo: an ML model trained on Latvian real estate market data that forecasts property values, applied to mortgage collateral assessment and scoring.

    fig. 09 — Verteo across Latvia
    esc
    ↑↓ navigate ↵ open