AI Solutions: knowledge continuity
Client work
Knowledge continuity pilots
7
week pilot: three weeks to set up, four to prove it.
3countries, from electronics retail to manufacturing
Status as of September 2026. Client names available under NDA.
Selected work
What we built, and what we would not claim.
Every case below carries its source and its date. Where a number does not exist we say so rather than reaching for one, and every page ends with what we are not claiming. If you would rather talk it through, book a 15-minute call.
Client work
Work we were paid to do, for a named or described client.
Marketing: AEO and GEO
Client work
Keyper
90%
of 73 tracked buyer questions name the brand in AI answers.
271AI citations in one month, across 69 pages
7AI platforms sending real visitors
Figures from a June 2026 reading, published September 2026.
Read the case studyMarketing: content and AEO
Client work
Allied Bank
67%
of tracked AI prompts cite the bank.
Figures from an August 2026 prompt panel, shown as proportions at the client's request.
Read the case studyMarketing: AEO and GEO
Client work
Hulii
8
collections restructured with redirects, keeping their search equity.
3schema types across the storefront: Product, FAQ, BlogPosting
llms.txtpublished so model readers can parse the store
Work delivered July and August 2026, published with the client's sign-off.
Read the case studyAI Solutions: AI application
Client work
Valoov
4
valuation methods run on every submission, DCF included.
2languages of tax documents read automatically
Liveat valoov.ai, built on Google Cloud
Client engagement. Status as of September 2026.
Read the case studyMarketing: content operations
Client work
Regional property portal
Thousands
of human-written articles produced through one content workflow.
Humanwritten, with automation deciding where effort lands
Auditedat portal scale, tracked through to a fix
Ongoing engagement. Client name available on request.
Read the case studyEngineering: operations systems
Client work
Client operating systems
5
operations run on one platform, from payroll to fuel.
USexam-management platform, built and maintained
Recurring engagements. Client names available on request.
Engineering: public sector web and GIS
Client work
Karachi Water and Sewerage Corporation
1 hour
response on a fatal issue, under a twelve-month SLA.
4severity tiers, tracked on a shared board
Government of Sindh entity. Engagement details available on request.
Marketing: AEO for a local firm
Client work
Barrett Crime Law
32
pages, built to be quoted rather than to cover everything.
3pillar clusters, rather than a sprawling site
llms.txtpublished alongside LegalService and FAQ schema
Built July and August 2026. Not deployed yet, so there are no results.
Read the case studyEngineering: operations systems
Client work
Municipal sanitation operations
3
pay models computed at once: per shift, per tonne and per trip.
Datedrate standards, so history does not rewrite itself
Weighedvehicle movements behind every tonnage payment
System described from its codebase, September 2026.
Read the case studyOur product
Software we build, run and operate ourselves.
AI: agent-operable software
Our product
Vivotics
MCP
server, so the workspace is a tool an assistant can operate.
1command to connect Claude Code or Claude Desktop
Humanapproval before any meeting output becomes a task
Live in production since July 2026. Our own platform, not sold as a product yet.
Read the case studyAI Solutions: agent fleet
Our product
COSMOS
19
operations run per brand, from research to ads and finance.
1QA agent reviews every artifact before it ships
0ad spend switched on without a person
Status as of September 2026. Client names available on request.
Read the case studyAI Solutions: AI operations
Our product
Our AI operations platform
25
client workspaces, each with its own knowledge base.
1tenant per client, so no data crosses over
Status as of September 2026.
Engineering: warehouse and retail
Our product
Cubitrek WMS
1,000+
units a month processed for a Norwegian electronics retailer.
Livelogistics link, verified on the partner's own system
2countries of retail point of sale for Brand Merchant
Status as of September 2026. Built on a licensed commercial core we have extended.
Read the case studyEngineering: web and mobile marketplace
Our product
LikeSpaces
344
live listings, counted from the platform's own API.
2native apps, live on iOS and Android
2countries served from one codebase
Listings counted 17 September 2026. We publish the count, not a rounded-up figure.
Read the case studyTeam track record
Work our people did in earlier roles. Described by sector, never named.
Cloud and Security: AI infrastructure
Team track record
Digital marketplace platform
~1B
requests a day, including GPUs for a multi-model LLM platform.
~40%cloud and spot spend cut
800+Kubernetes clusters in scope
Delivered by our infrastructure lead in a previous role. The client is not named.
Cloud and Security: cloud architecture
Team track record
Grocery retail group
30%
less in-store theft, reported for the video analytics program.
40+cloud-native apps built to one architecture
Delivered by our infrastructure lead in a previous role. The client is not named.
Cloud and Security: AI observability
Team track record
Connected-vehicle platform
85%
cut in mean time to diagnose across the platform.
22M+connected vehicles supported
100%SLA maintained across platform operations
Delivered by our infrastructure lead in a previous role. The client is not named.
How to read these
Three kinds of evidence, labelled.
Client work, our own products and our team's earlier track record are different things and we do not mix them. A product we built and run ourselves is real evidence that we ship, but it is not a client result, so it never sits on the page pretending to be one.
Want your own baseline?
We measure before we start, so there is something to measure against.
Most of the work above began with a reading taken before anything changed. That is the part that makes a result mean something later.