Skip to content

06 · AI Development

AI that does real work inside your product

We design, train, and ship AI that earns its place in production — from custom models to generative systems — backed by data pipelines and monitoring that keep it reliable at scale.

Request a quote What we deliver

You own the code and IP Weekly demos Reply within one business day

AI

Copilot · fine-tuned on your data

Live
Which invoices are overdue by more than 30 days?
14 invoices totalling 42,300 are overdue. Three accounts carry 68% of it — I’ve drafted reminders for review.

0.94

F1 score

86ms

Inference

12k/d

Requests

AI Development, in short

OlDevs is a Vancouver technology studio that designs, trains, and deploys custom AI systems for corporations, associations, government bodies, franchises, and startups.The service covers custom machine learning models, generative AI and LLM applications such as chatbots, copilots, and RAG systems, natural language processing, computer vision, AI agents and workflow automation, and predictive analytics. Every system ships with data pipelines, evaluation, and monitoring so it stays accurate in production, and the client owns the models, code, and data.

Key facts

01Service
Custom AI & ML development
02Focus areas
LLMs, NLP, vision, agents, forecasting
03Delivery
Prototype to production MLOps
04Hosting
Cloud, on-premise, or hybrid
05Ownership
You own models, code, data
06Studio
Vancouver, BC — since 2014

06 · AI Development

Intelligence, built into the product.

Six disciplines, one system

01

Custom ML Solutions

Bespoke models trained on your data, taken from prototype to production-grade MLOps.

Model training · MLOps · Pipelines · Deployment · Tuning

Explore02

Generative AI & LLMs

Chatbots, copilots, and RAG systems on leading language models, fine-tuned to your domain.

LLMs · RAG · Fine-tuning · Copilots · Prompting

Explore03

Natural Language

Classification, sentiment, entity extraction, semantic search, and translation.

Classification · Sentiment · NER · Search · Translation

Explore04

Computer Vision

Image & video recognition, object detection, OCR, and automated quality inspection.

Detection · OCR · Recognition · Segmentation · Video

Explore05

AI Automation & Agents

Autonomous agents and workflow automation that remove repetitive, manual work.

Agents · Workflow · RPA · Orchestration · APIs

Explore06

Predictive Analytics

Forecasting, recommendation engines, and decision dashboards drawn from your data.

Forecasting · Recommenders · Scoring · Dashboards

Explore

By the numbers

Where AI project effort actually goes.

The model is rarely the hard part.

Data readiness35%

Access, cleaning, labelling, pipelines

Integration & UX25%

Wiring the model into real workflows

Modelling20%

Training, fine-tuning, prompting

Evaluation & guardrails15%

Test sets, monitoring, safety

Deployment5%

Serving, scaling, handover

Illustrative shares of a typical engagement — every project is scoped individually.

Process

How we build it — from data to deployment

  1. Discover

    We map the decision or task the AI should own, agree on success metrics, and confirm the idea is worth building before anyone trains anything.

  2. Data

    We audit, clean, label, and pipeline the data you already have — and tell you plainly if it is not yet enough.

  3. Model

    We prototype quickly, compare approaches against a shared evaluation set, and harden the winner for production.

  4. Integrate

    The model is wrapped in APIs, connected to your apps and workflows, and tested with the people who will actually use it.

  5. Monitor

    Once live, we track accuracy, drift, latency, and cost, retraining and tuning on a schedule that keeps results dependable.

How we work

How we make AI production-grade

01

Data audits before modelling

We profile, clean, and label your data first, because a model is only as good as what it learns from.

02

Evaluation sets, not guesswork

Every model and prompt is scored against a held-out test set, so improvements are measured rather than assumed.

03

Retrieval before retraining

For LLM work we ground answers in your own documents with RAG first, and fine-tune only when the evidence shows it pays off.

04

Human-in-the-loop by design

Confidence thresholds route uncertain cases to people, so automation speeds work up without hiding its mistakes.

05

Versioned data, features, and models

Datasets, feature code, and model weights are versioned together, so any prediction can be traced and reproduced.

06

Drift and cost monitoring

Dashboards watch accuracy, latency, and token spend in production and alert your team before quality slips.

Who it's for

AI Development for organisations that have to get it right.

Whether the audience is a customer, a member, a citizen, or your own team, the work has to hold up under real use.

Corporations

Forecasting, document intelligence, and copilots that sit inside the systems your teams already use, with the governance your IT, legal, and risk groups expect.

Associations & government

Member and citizen services that answer faster — search, triage, summarisation, and translation built with privacy, accessibility, and Canadian data residency in mind.

Franchises

One model, every location: demand forecasting, quality inspection, and support automation that give head office a consistent picture across the whole network.

Entrepreneurs & startups

An AI-first product built properly the first time — a working prototype fast, then the pipelines and monitoring your investors and customers will ask about.

12+

years of engineering experience

1 business day

to reply to your quote request

40%

average cut in manual processing time

Works well with

Related services

All services
05

Full-Stack Development

End-to-end architecture, APIs, databases, and DevOps.

Explore01

Web App Development

Custom web applications, dashboards, and SaaS platforms.

Explore07

Performance Marketing

SEO, paid media, content, social, and analytics — the growth engine behind every build.

Explore

Selected work

Selected AI development work.

Client names are withheld; the problem and the outcome are the point.

01AI

A national commercial insurance brokerage with around 450 staff across six offices

Industry
Insurance & financial services
Services
AI development · Generative AI & LLMs · Full-stack development
Duration
14 weeks
“We were nervous about an AI giving confident wrong answers. What we got is a tool that says it cannot find something when it should, and shows its work when it can.”
— Chief Compliance Officer, national insurance brokerage

Challenge

Brokers answered policy-wording questions by searching a shared drive of 30,000 PDF wordings, endorsements and carrier bulletins, or by asking whichever senior colleague happened to be free. New hires took months to become confident, and the compliance team had no way to know which document version a broker had relied on when advising a client.

Solution

We built an internal assistant on a large language model that answers questions only from the brokerage's own document library and shows the exact passage and version it drew from. The retrieval layer indexes wordings, endorsements and bulletins as they land, keeps carrier-specific permissions intact, and declines to answer when it cannot find support in the source material. It runs inside the brokerage's own cloud tenancy, so no client or carrier data leaves their environment, and an audit log records every question, answer and citation. The full-stack build covers the ingestion pipeline, the broker-facing web app and the compliance review screens.

78%

Of wording questions resolved without escalating to a senior broker

11 min → 40 s

Median time to a cited answer

100%

Of answers traceable to a document version

Read the case study

02AI

A two-founder agritech startup in the Fraser Valley

Industry
Agriculture technology
Services
AI development · Computer vision · Web app development
Duration
12 weeks
“Investors did not ask to see a deck. They asked to see the greenhouse.”
— Co-founder & CEO, agritech startup

Challenge

The founders had a hypothesis that early signs of plant stress in greenhouse crops could be spotted from ordinary cameras days before growers noticed them, and a pilot grower willing to try. What they did not have was a working model, a way to show growers the results, or anything a seed investor could put their hands on.

Solution

We started with the data: a labelling workflow the pilot grower's staff could use on a tablet during their normal walk-throughs, which produced a training set in three weeks. On top of it we trained a computer-vision model to flag stress indicators from fixed camera feeds and shipped it with a web app that shows growers a heat map by bay and a daily digest of what changed. Weekly demos ran in the greenhouse, not a boardroom. The founders own the model weights, training pipeline and code outright, and used the working pilot to close their seed round.

5 days

Average lead time on stress detection versus manual scouting

0.91

Precision on held-out pilot data

3 weeks

From first camera install to first flagged bay

Read the case study

03AI

A provincial electric utility in Manitoba with roughly 1,200 field and office staff

Industry
Energy and utilities
Services
Generative AI & LLMs · Natural Language · AI Development
Duration
5 months
“Our crews stopped phoning the control centre for procedure questions, and when the copilot answers they can see exactly which bulletin it came from.”
— Manager of operational readiness, provincial electric utility

Challenge

Operating procedures, safety bulletins and collective agreement clauses lived across SharePoint, a document management system and two decades of PDFs, so answering a straightforward question often meant asking a colleague who had been there long enough to remember. Crews in the field gave up and phoned the control centre instead. Anything an assistant said had to be traceable to the document it came from, and staff could only ever see the documents their role already permitted.

Solution

OlDevs built a retrieval system over the utility's own document set, chunked and indexed in PostgreSQL with pgvector, with answers generated by a large language model inside the utility's own tenancy so no content left their boundary. Every response cites the source document, section and revision date, and links straight to it. Permissions are read from existing Microsoft Entra ID groups at query time, so the copilot retrieves only what the person asking is already cleared to open.

6 wk

To first department live

−63%

Time spent locating policy answers

+41%

Weekly active internal users

Read the case study

FAQ

AI Development — questions we hear first.

A focused proof of concept typically takes four to eight weeks, and a production-ready system with pipelines, integration, and monitoring usually takes three to six months. We confirm the timeline during the Discover phase once we have seen your data and the systems it needs to connect to.

You do. Trained model weights, source code, prompts, pipelines, and the data they run on are assigned to your organisation on completion, and we never reuse your data to train work for anyone else.

We build custom models in Python with PyTorch and scikit-learn; generative systems on OpenAI, Anthropic, and Google models as well as open-weight models such as Llama and Mistral; and deploy on AWS, Azure, Google Cloud, or your own servers. We choose the stack to fit your data, budget, and compliance needs rather than the other way round.

Yes. Our support plans cover monitoring for accuracy drift, scheduled retraining, prompt and model upgrades as providers change, and a named engineer your team can reach. We can also hand the whole system over to your in-house team whenever you are ready to run it yourselves.

Data is processed in Canadian or client-chosen regions, encrypted in transit and at rest, and shared with third-party model providers only under terms that prohibit training on it — or not at all if you choose a private or on-premise deployment. We work within PIPEDA, BC's FIPPA, and your own governance and audit requirements.

We plug into your existing tools and rhythms — a shared repository, weekly demos, and a channel your people can post in — and we pair with your developers and analysts so the knowledge stays in-house. If you already have data scientists, we can focus on the engineering, integration, and deployment work around them.

It depends on the data you have, the complexity of the models, and how deeply they need to integrate with your systems, so we scope every project individually rather than publish a rate card. Request a quote with a short description of the problem and we will reply within one business day with a recommended approach and fixed-scope options.

Still have a question? Ask us when you request a quote

Let’s connect

Let’s scope your AI development project.

Tell us what you’re building. We’ll reply within one business day with next steps and a tailored quote — no obligation.

We’ll only use your details to prepare your quote. No lists, no spam.

Call us Request a quote