Skip to content

Engineering, cloud and support

+44 20 4604 8374
Hexploits

Applied AI

Internal AI assistants and copilots for UK businesses

An internal AI assistant is a system that answers staff questions and drafts work from your own documents, tickets and records, rather than from general internet knowledge. Hexploits builds assistants that retrieve from your systems under your existing permissions, log every answer with its sources, and run inside your cloud account or on EU-hosted infrastructure.

Request a proposalTalk to an engineer

A named engineer replies within one working day. A written scope and an indicative price within two.

Who this is for

  • Operations and support teams answering the same questions from the same documents every day.
  • IT directors asked for a company-wide AI assistant that has to respect existing access control.
  • Professional services firms whose know-how lives in past reports, proposals and emails.

What you get

Deliverables, not slogans. Each one appears in the statement of work.

  • A working assistant integrated with your document stores, ticketing and line-of-business systems.
  • Retrieval that honours existing permissions: a user only sees answers built from documents they can already open.
  • Source citations on every answer, and a log of questions, answers and sources for audit.
  • An evaluation set built from your real questions, with measured answer quality before and after launch.
  • Deployment in your AWS, GCP or Azure account, or on Hexploits Cloud in the EU.

Our engineers work across the major languages, frameworks and cloud platforms. We build on the stack you already run, with technology choices explained in writing before work begins.

How it is delivered

The same four stages as every Hexploits engagement, applied to this capability.

  1. Stage 1

    Discovery

    Two to four weeks. We identify the questions people actually ask, the systems that hold the answers, and the permissions model. Output: a written scope, an evaluation set and a cost model.

  2. Stage 2

    Build

    Ingestion and indexing of your sources, retrieval tuned against the evaluation set, and the interface people will use, usually inside Teams, Slack or the intranet.

  3. Stage 3

    Launch

    Security review, a pilot group, and a measured comparison against the baseline before wider rollout.

  4. Stage 4

    Operate

    Ongoing re-indexing, model updates, monitoring of answer quality and cost, and a monthly report under managed support.

How success is measured

Every engagement agrees its measures and the measurement period in writing before work starts.

  • Time to answer internal questions, measured against a baseline sample before launch.
  • Deflection of tickets that no longer reach a person.
  • Answer accuracy against the evaluation set, reported monthly.
  • Cost per query, so the finance director can see the bill before the rollout.

Proof

Case studies with numbers, and reviews linked to Google where they were left there.

98%Faster page loads

Gradvisor · Charity · 8 weeks to production, then ongoing

98% faster page loads and a 12% smaller cloud bill for a UK careers charity

12% lower cloud cost · 2 days to ship a change
75%Under the original budget

swarmd.ai · Software vendor · 6 months

Enterprise AI control plane delivered in six months at 75% under budget for a UK software vendor

75% ahead of the original timeline · 6 months from blank page to paying enterprise customers

  • Director, Gradvisor

    “Fantastic company and our development partner for Gradvisor, a social mobility careers platform with national ambitions. Extremely responsive and mission-oriented. Cameron owns any shortfalls humbly - rare for IT providers. Thinks like a client too.”

    Tushar PrabhuDirector, GradvisorRead the review

  • Director, JobVantage

    “Working with Hexploits has genuinely been a pleasure, and I see them as my scaling partner for the foreseeable future as JobVantage grows. If you’re looking for a development team who combine strong AI/engineering capability with honesty, flexibility and a real interest in your business, I’d strongly recommend them.”

    Brandon BowdenDirector, JobVantageRead the review

Questions we get asked

Will the assistant see documents a user is not allowed to see?
No. Retrieval is filtered by the permissions the user already holds in the source system, so an answer can only be built from documents they could open themselves.
Does our data train the model?
No. We use enterprise API terms that exclude training on inputs, or private open-weight models where nothing leaves your boundary, and we document which applies.
How do you stop it making things up?
Every answer is grounded in retrieved documents and cites them. Questions with no supporting source get a "not found" rather than a guess, and the evaluation set measures how often that happens.
What does it cost to run?
It depends on query volume and model choice. Discovery produces a cost model per query and per month before anything is built, and managed support reports the actual figure monthly.

Related

Sectors where this is most often needed

Reading

More in applied ai

Applied AI overview

Next step

Request a proposal.

Tell us about the system and the sector. A named engineer replies within one working day. A written scope and an indicative price within two working days of a short scoping call.

Request a proposalTalk to an engineer