Skip to content
dhii.ai

Model Training & ML

Custom Model Training (Vision, NLP & Tabular)

Purpose-built models for one job, trained on your data — often smaller, faster and more accurate than a general-purpose LLM for that job.

Data audit
1–2 weeks
First production model
4–10 weeks
Typical team
1–2 ML engineers

The right tool for a narrow job

Large language models get the headlines, but plenty of valuable problems are better solved by a dedicated model trained for one purpose: spotting a defect in a product photo, routing a document to the right department, predicting which invoices will be paid late, flagging a suspicious transaction. These models are small enough to run cheaply at high volume, consistent enough to audit, and often more accurate than a general model prompted to do the same thing.

Framing comes before modelling

Many ML projects fail before any code is written, because the question is vague. “Predict churn” becomes useful only when it says which customers, how far ahead, and what action the prediction will trigger. We spend the first days framing the target, the decision it supports and the metric that matters to the business — which is rarely raw accuracy.

Honest validation

The most common ML mistake is a model that looks great in testing and fails in the real world, usually because of data leakage or a test set that doesn’t resemble production. We split data the way the model will be used (often by time), check for features that “cheat”, and compare every model with a simple baseline. If a gradient-boosted tree beats a neural network, we ship the tree.

Built to run

A model is only useful in production. We package models as versioned services or batch jobs with input validation, latency budgets and logging, and we set up monitoring for data drift and performance decay. Each model ships with a model card describing what it was trained on, where it performs well and where it shouldn’t be trusted.

Deliverables

What you get

  • 01

    Data audit

    Assessment of volume, quality, labels, leakage risks and bias, with a go/no-go recommendation.

  • 02

    Problem framing

    A precise definition of the prediction target and the business metric it serves.

  • 03

    Baseline and experiments

    Simple baselines first, then progressively stronger models, all tracked and reproducible.

  • 04

    Trained model

    The selected model with its training code, weights, and a model card describing limits.

  • 05

    Inference service

    A containerised API or batch job, with input validation and monitoring hooks.

  • 06

    Monitoring plan

    Drift detection and a retraining schedule tied to real performance.

Process

How a custom model training (vision, nlp & tabular) project runs

  1. Step 1

    Frame

    Turn the business question into a measurable prediction task.

  2. Step 2

    Audit data

    Check what data exists, how it's labelled and what's missing.

  3. Step 3

    Experiment

    Baselines, feature work and model selection with honest validation.

  4. Step 4

    Productionise

    Package, test and deploy the model with monitoring.

  5. Step 5

    Iterate

    Retrain as data grows and monitor for drift.

Engagement models

Ways to work together

Every engagement starts with a fixed, written quote. See engagement models & pricing.

FAQ

Custom Model Training (Vision, NLP & Tabular): common questions

Why train a custom model when LLMs can do so much?

For a narrow, high-volume task — say classifying millions of transactions or images — a small custom model is usually faster, cheaper, more consistent and easier to explain than an LLM. LLMs remain useful for labelling data and handling the long tail.

How much labelled data do we need?

It varies widely. Fine-tuning a pre-trained model may need only hundreds of labelled examples per class; training tabular models depends on how predictable the target is. The data audit gives a realistic answer for your case.

Can you help label our data?

Yes. We set up labelling workflows, write guidelines, and use model-assisted labelling to speed up human reviewers. See our synthetic data and labelling service.

Do we own the model and code?

Yes. Training code, weights, data pipelines and documentation are delivered to you.

Related services

Start a project

Have a workflow that should run itself?

Tell us what you want automated or built. You get a written assessment of feasibility, risks and a fixed quote — usually within two business days.