Chrysalis Core Email us

Cape Town, South Africa · UTC+2 · on-site or remote

AI pilot to production

We teach your people what AI can and cannot do, get your company ready to run it safely, and put our engineers inside your organisation to build what holds up.

  • TrainingLeaders and teams, on their own work
  • AI EnablementData, tests and access in place before AI goes live
  • Embedded AIOur engineers in your organisation, from first interview to production
Email us

or copy the address michaelm@chrysaliscore.com

What we do

Training

Executives and teams learn what AI can and cannot do, on their own documents and decisions rather than demos.

You get

  • For leaders: the uses worth funding, the check each one needs, and the ones to drop.
  • For teams: practice at catching an answer the assistant made up, before it reaches a customer.
Example · a practice round on your own documents
Your team asked

Can we pay this supplier early and take the discount?

The question comes from their own work, not from a demo.

The assistant answered

Yes. Your payment policy allows early payment to any supplier.

It sounds certain and names your own policy. It is wrong.

The check

The payment policy says nothing about early payment. The assistant made the rule up.

Anyone on the team can do this: open the source the answer names.

What the team keeps: before acting on an answer, open the source it cites.

AI Enablement

We get your company ready to run AI safely: what it may read, what it may never touch, and the tests every change must pass before it goes live.

You get

  • A list of what AI may read, agreed with the team that owns each source, with personal details hidden. Everything else stays out of reach.
  • Read-only access unless a change is approved.
  • A test set of your own questions, rerun on every change.
  • A priority list your sponsors can see, with what each item displaces.
Example · a question the assistant must refuse
Someone asked

What does the finance manager earn?

Someone will ask. What matters is what the assistant can reach when they do.

The assistant answered

Sorry, payroll is not available here. HR can help with that.

It refuses because payroll was never connected. No instruction has to hold.

What it can read

Open invoices · Contract text · Support tickets

What it can change

Nothing

The data owners agreed this list. Personal details stay hidden even here.

Tested

This question is in the test set. It is asked again after every change.

A fixed set of your questions with known right answers. A change that breaks one does not go live.

What you keep: an assistant that sees only what you approved, and a test that shows it still behaves after every change.

Embedded AI

Our engineers work inside your organisation: they interview the people involved, build a prototype on your data in days, and take only what held up into production.

You get

  • Engineers in your organisation, on-site or remote, from the first interview.
  • A prototype on your real data in days, and the test that says whether it held.
  • A ranked list of opportunities for your sponsors to put in order.
  • The production system, its tests, a named owner on your side, and the runbook.
Example · one request, from first interview to production
  1. Listen

    “Can AI answer our complaints?” The problem underneath is often narrower: the same kinds of complaint, answered from scratch each time.

  2. Prototype

    Drafts for recent complaints, set beside the replies your team actually sent.

  3. Build for production

    What then runs inside your organisation:

Complaint

Charged twice for one order.

Draft reply

Sorry about that. The second charge will go back to your card. Your order is not affected.

Drafted, never sent by the system. A person on your team sends every reply.

Source

Refund policy: duplicate payments

Status

Draft · waiting for a person

Each draft names the rule it used, so the person sending it can check.

Not drafted

Refund disputes go to your team as before.

Only the kinds of complaint that passed the prototype. The rest stay with people.

What you keep: a working system inside your organisation, its tests, a named owner and the runbook.

How an engagement runs

We listen first, test ideas on your real data in days, and build for production only what held up.

  1. Listen

    We interview the people closest to the strategy and the people closest to the work, and trace each request to the problem behind it.

    You receiveA written list of the AI opportunities, ranked, with the expected value of each and what it depends on.

  2. Prototype

    A small experiment on your real data answers the most expensive question first, in days rather than months.

    You receiveA working prototype on real data, and the evaluation that says whether it held.

  3. Build for production

    Only an idea that passed its prototype is built for production, and you see its progress throughout.

    You receiveThe governed system, its regression harness, a named owner on your side, and the runbook.

The rules we work by

How we deliver

  • DiscoveryEvery request is traced to the problem behind it before anything is built.
  • PrototypesPrototypes are built fast. Production systems are built deliberately.
  • ProofA proof of concept that never ships can still pay for itself.
  • VisibilityProgress is visible to the people paying for the work.
  • AdoptionWe stay approachable and take people’s concerns about AI seriously.

What we will not do

  • AccessWe never give AI unrestricted access to business-critical systems or data.
  • DataWe do not build on ungoverned data. We work from governed sources, with the teams that own them.
  • PrioritiesNo request is taken on without saying what it displaces. Priorities are explicit and agreed.
  • PromotionNo prototype reaches production before it has passed its evaluation.

An engineering company

Our engineers come from healthcare, banking, automotive and logistics. They have modernised legacy platforms, built event-driven systems on AWS and GCP, and carried the budgets and vendor relationships that come with production software.

  • On your projectThe same engineers do the whole job: they interview your people, build the prototype and the tests it must pass, write the production code, and hand the system to a named owner on your side with its runbook.
  • InfrastructureThey also set up what the system runs on: least-privilege access, DNS, TLS, VPN tunnels, firewalls and monitoring.
  • Your emailAn engineer reads it and replies, one of the engineers who would work on your project.

Start with one email

Tell us what you want AI to do for your company.

Pick the request closest to yours, and the email below fills in to match.
Training
AI Enablement
Embedded AI

Tomichaelm@chrysaliscore.com

SubjectAI idea: Show our executives what AI can do

Hello Chrysalis Core,

We would like help to: show our executives what AI can do

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhich decisions do your executives want help with, and what do they already believe AI can do?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI idea: Teach our teams to check AI’s work

Hello Chrysalis Core,

We would like help to: teach our teams to check AI’s work

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhere does your team already use AI, and who checks its answers today?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI idea: Let AI read our data safely

Hello Chrysalis Core,

We would like AI to: read our data safely

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhich questions do people ask of your data, and which sources must AI never read?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI idea: Test our assistant before it goes live

Hello Chrysalis Core,

We would like help to: test our assistant before it goes live

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhat must the assistant always get right, and how would you know today if a change broke it?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI idea: Answer customer complaints

Hello Chrysalis Core,

We would like AI to: answer customer complaints

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhich complaints repeat, and what does a wrong reply cost?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI idea: Summarise our contracts

Hello Chrysalis Core,

We would like AI to: summarise our contracts

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWho reads your contracts today, what are they looking for, and which clauses carry the risk?

Open this email

Tomichaelm@chrysaliscore.com

SubjectAI at our company

Hello Chrysalis Core,

What we are trying to do:

Where our AI work is today:

The first thing we would ask youWhat problem sits behind the request, and who feels it today?

Open this email

or copy the address michaelm@chrysaliscore.com

What happens after you email

  1. An engineer reads your email and replies.
  2. If your email does not already say, the reply asks two things: what you are trying to do, and where your AI work is today.
  3. It proposes a first conversation.