khak.ai Intelligence. Made useful.

Your team shouldn't be retyping things.

khak.ai builds AI and workflow automation that moves information from where it arrives to where it's needed — document automation, reconciliation, filing and reporting — with a person approving every step. For charities and growing businesses, wherever you are.

Book a 30-minute call See the work

What clients stopped retyping

Bank statementsReconciled ledger
Photos of booksCatalogue records
Shipping PDFsCustoms entries
Lecture recordingsFinished manuscripts

A person approves every step. That isn't a setting you can turn off.

Most organisations have now been burned by something an AI tool got confidently wrong. Everything built here drafts, and then waits: nothing is submitted, filed, sent or posted until someone on your team has looked at it and said yes. The speed comes from removing the typing, not from removing the judgement.

Your systems, your accounts

Built to run on the tools and infrastructure you already pay for. You're told which model sees what, and what is kept out of it, before anything is built.

An honest verdict first

Some processes shouldn't be automated at all. If yours is one of them, that's what the audit will say — in writing, before you've committed to a build.

01 — WHAT I BUILD Five kinds of problem, across finance, operations, compliance and publishing.

Document & data automation

Structured data pulled out of PDFs, scanned forms and web portals, so nobody is rekeying the same fields into a second system.

Financial reconciliation & reporting

Payment processors, bank feeds and ledgers matched into one audit-ready export, built for the accounting system you already use.

Regulatory & portal filing

Form-heavy submissions to tax, customs and compliance portals, drafted end to end with a mandatory human approval before anything is filed.

Custom internal tools

Small, purpose-built assistants trained on your own documents, policies and house style — rather than a general chatbot bolted onto the side.

Knowledge & content pipelines

Transcription, translation and publishing pipelines that turn raw recordings and drafts into finished, reviewable output in hours.

02 — SELECTED WORK Three projects, told plainly: the problem, what I built, what changed.
01  ·  Charity finance

Nine payment providers, one reconciled ledger

The problem

Donations arrived through nine separate processors — card readers, direct debit, online gateways — each with its own fee structure and naming. Finance matched them to the bank by hand, every month.

What I built

An engine that reads every provider export and bank statement, matches the lines, separates fees and writes accounting-ready files. Gift Aid claims draw from the same cleaned donor records.

What changed

Nine providers and the bank reconciled into one export, with fees separated automatically.

1,845 donations and 590 donor records cleaned and linked to claim sheets.

02  ·  Specialist library

Cataloguing by photograph

The problem

Every new acquisition was typed in by hand, including Arabic titles and transliteration, so the backlog only ever grew.

What I built

The librarian photographs the title page. The system drafts a catalogue record using the library's own cataloguing rules, and the librarian approves or corrects it in the library system.

What changed

10× faster cataloguing, with every record still checked by a person.

The backlog stopped growing faster than it could be cleared.

03  ·  Clearing & warehousing

From a paper stock book to a live platform

The problem

Stock lived in a handwritten book. Partial releases and late notice of collections led to miscounts, shrinkage and disputes with clients over what was actually held.

What I built

A stock platform tracking every lot by package, pallet, carton or bale, with a running balance per client — plus a clearing workflow that turns shipping documents into draft customs entries for approval.

What changed

A handwritten ledger replaced by a per-client balance anyone can check.

Built and in rollout; customs entries are drafted automatically and approved before submission.

03 — MORE WORK Publishing, logistics, finance and one thing just for fun.
ProjectAreaWhat changed
Lecture-to-book pipeline Publishing Recorded lectures become publication-ready manuscripts in the house style. Months of editing down to hours.
Lecture dubbing & translation Publishing Specialist lectures transcribed, translated against a controlled glossary and narrated in English, synced to the original recording.
House-style copy-editor Publishing Staff self-edit to the publisher's own style guide, replacing £1–5k of external copy-editing per work.
Donor records & tax claims Charity finance Donation and donor records cleaned, deduplicated and linked to claim sheets. Card-reader gifts coded at source.
Customs documentation Logistics Shipping PDFs read and drafted into portal entries. A person approves before any submission.
Internal assistants Operations Custom assistants for finance, HR and teaching teams, each trained on the organisation's own policies.
Anaan Lab An animated series and storybooks: consistent characters, narration and music from one production pipeline.
04 — HOW I WORK Fixed scope, fixed fee, and a verdict before a build.

Audit

1–2 weeks

I sit with one process, map where information gets retyped, and put a number on the time it costs. You get a plain verdict: build, fix the process, or leave it.

Build

3–6 weeks

A fixed scope at a fixed fee. Working software in small steps, with a person approving every output before it counts.

Care

Monthly

I keep it running, fix what breaks and make small improvements as your work changes.

05 — QUESTIONS The five things people ask before the first call.
How do you stop AI getting things wrong?

Nothing is submitted, sent or filed without a person approving it. The automation drafts; your team confirms. Every workflow is built with that approval step in it, and it is not optional.

What does an audit cost, and what do I get?

An audit runs one to two weeks at a fixed fee. You get a map of where information is being retyped, a number on what it costs you, and a plain verdict: build it, fix the process instead, or leave it alone. The audit stands on its own — you are not committed to a build at the end of it.

Do you work with organisations outside the UK?

Yes. The work is remote, and clients span the UK, Europe and East Africa. Working hours overlap UK and European business hours, and nothing about the delivery depends on being in the same room.

What happens to our data?

Systems are built to run on your own infrastructure and accounts wherever possible. Where a model provider is involved, you are told which one, what it sees, and what is excluded — before anything is built, not after.

What if automation is the wrong answer for us?

Then the audit says so. Some processes are too low-volume, too judgement-heavy or too risky to hand to a model, and the better fix is a cleaner process or a better form. You get that verdict in writing rather than a project you didn't need.

06 — ABOUT

Built by someone who ran the processes first.

khak.ai is the practice of Ali Khaki. For years I ran operations and finance systems for a UK charity: donations, tax claims, the library, the building. Most of what slowed us down wasn't hard work. It was the same information being typed in twice.

What's offered here is what I learned fixing that — built by someone who has actually run the finance, donor and admin processes the software is meant to replace, rather than someone who has only read about them.

khāk is Persian for earth. The unglamorous groundwork that makes the clever part hold.

BackgroundOperations and digital transformation, charity and logistics sectors
EducationMSc Computer Science, University of Birmingham
Working hoursUK and European business hours
ClientsUK, Europe and East Africa

What does your team retype every week?

Tell me in a 30-minute call. No pitch, no obligation — and if automation isn't the answer, I'll say so on the call rather than sell you a project.

Book a 30-minute call hello@khak.ai