Skip to content
SUPPORT SYSTEM
AI & Automation

Automation that removes the work nobody wanted

We automate the repetitive parts of a business process and, where it genuinely helps, put a language model behind a step that used to need a person reading something. Always inside software you control, with the decisions auditable.

What we do, and what we do not claim

Vertex Arc does not train foundation models and does not own one. What we do is integrate established model APIs — from providers like OpenAI and Anthropic — into products where they earn their place, and build the far less glamorous automation around them that actually removes hours from a week. A lot of the value is not the model at all. It is the routing, the queueing, the validation, the retry, the human review step, and the audit record of what was decided and on what basis. A model that classifies an invoice is a component; the system that receives it, extracts it, checks it against a purchase order and escalates the exception is the product. We are also clear about where this does not belong. If a rule can be written down, write the rule: it is cheaper, faster, testable and it does not change its mind. We will say when that is the better answer, and often it is.

Key benefits

What this changes for your business.

Hours back, measurably

We pick processes where the time saved can be counted before the work starts, so the result is not a matter of opinion.

A person stays in the loop where it matters

Automated steps that carry risk route to a human with the evidence attached, rather than deciding silently.

Your own content, actually findable

Semantic search over documents, tickets and knowledge bases, answering from your material rather than from the open web.

Your data boundary is explicit

What is sent to a model provider, what is retained and what never leaves your infrastructure is written down before anything is built.

What we deliver

The things you actually receive.

  • Document processing

    Extracting structured data from invoices, contracts and forms, with confidence thresholds and a review queue.

  • Intelligent search and Q&A

    Search across your own documents that understands the question, with citations back to the source.

  • Classification and routing

    Tickets, emails and requests sorted to the right queue with the right priority, and a fallback when confidence is low.

  • Drafting assistants

    First drafts of replies, summaries and reports inside your own tools, with a person approving before anything is sent.

  • Workflow automation

    The unglamorous part: scheduled jobs, event triggers, approvals and integrations that remove a manual handover.

  • Internal productivity tools

    Small applications that take a recurring internal task and reduce it to one screen.

Core capabilities

The engineering disciplines this service draws on.

Model API integration
Retrieval over your content
Process automation
Structured extraction
Human-in-the-loop review
Evaluation & guardrails
Cost & usage monitoring
Data boundary design
Audit trails for decisions

Technologies we use

The stack we would reach for, and what each part is for.

OpenAI API

A hosted model API for classification, extraction and drafting. Used as a component inside a product, with the business rules still ours.

Claude API

Anthropic’s model API, used the same way: a capable component behind our own validation, logging and access control.

Laravel

A mature PHP framework for secure, maintainable server-rendered applications and APIs, with authentication, queues and testing built in.

PHP

The language behind a large share of the web, and a fast, strictly typed one since PHP 8.

Node.js

A JavaScript runtime suited to real-time features and API gateways, where many connections spend most of their time waiting.

TypeScript

Static types over JavaScript. On a codebase several people maintain, it turns a class of runtime bugs into compile-time ones.

PostgreSQL

A relational database with strong support for JSON, full-text search and geospatial data, for models that outgrow plain tables.

Redis

An in-memory store used for caching, queues and rate limiting — the difference between a page that waits on the database and one that does not.

Elasticsearch

A search engine for when `LIKE %term%` stops being an answer — typo tolerance, relevance ranking and faceted filtering.

Technology adoption

Technologies in this stack are publicly documented as being used by organisations including those below.

Netflix

Node.js

Source

Slack

TypeScript

Source

These organisations are named as documented users of the technologies listed. They are not clients of Vertex Arc, and their inclusion does not imply any relationship with or endorsement of Vertex Arc.

Industries we serve

Sectors where this service tends to fit well.

  • Professional Services
  • Healthcare
  • Logistics
  • SaaS
  • Education
  • Media & Publishing

Our delivery process

How an engagement runs, from first conversation to ongoing support.

  1. Discovery

    We work out what the software has to do, who uses it, and which constraints are real. The output is a written scope, not a proposal.

  2. Architecture

    Data model, boundaries, integrations and infrastructure decided and agreed before anybody writes application code.

  3. Design

    Flows and interface, including the empty, error and permission states that decide how the product actually feels.

  4. Development

    Built in reviewable increments against a conventional structure, with tests around the parts that would be expensive to break.

  5. QA & security

    Functional testing, performance checks, and a review of authentication, authorisation and dependency risk before launch.

  6. Launch

    Deployment, monitoring, and a period of close attention while real traffic finds what staging did not.

  7. Continuous improvement

    Patches, upgrades and new work through the support system, so the product keeps being maintained rather than quietly ageing.

Use cases

What this looks like as a finished product.

Inbound request triage

Incoming emails and forms classified, summarised and routed, with the original always one click away.

Invoice and document intake

Line items extracted and matched against purchase orders, with only the mismatches reaching a person.

Internal knowledge search

Staff asking a question in plain language and getting an answer from your own policies, with the source cited.

Why Vertex Arc

We say when you do not need AI

If a deterministic rule solves it, we build the rule. It is cheaper to run, easier to test and it does not hallucinate.

No claims about models we did not build

We integrate established provider APIs. We do not describe them as our own technology, because they are not.

Cost is designed in

Token budgets, caching and model selection per task, so a feature does not become an unbounded monthly bill.

Auditable by design

Every automated decision records its input, its output and its confidence, so it can be reviewed rather than trusted.

Frequently asked questions

Does our data get used to train someone else’s model?
That depends on the provider and the plan, and it is a decision we make explicitly with you before anything is sent. The data boundary — what leaves your infrastructure, what is retained and for how long — is written into the design document.
What if the model gets it wrong?
It will, sometimes. That is why every design includes a confidence threshold, a review queue for low-confidence cases and an audit record. The question is not whether it errs but whether an error is caught before it costs anything.
Do you build custom models?
No. We integrate established model APIs and build the software around them. If a project genuinely needs a bespoke model, that is a different discipline and we will say so rather than take the work.