We build the data systems, and we check the AI that reads them.

We author and audit the tasks that frontier models are measured against. Current work includes auditing LLM text-to-SQL output for AfterQuery — judging whether generated SQL is correct against a real schema and the question's intent, and saying precisely where it fails. On Mercor's SWE-Agentic programme we authored agentic coding problems and graded model output against rubrics. For Turing we built Python services containerised with Docker against structured acceptance criteria, deterministic tests and reproducibility checks.

The discipline behind it is older than the AI industry. We designed the data quality assessment frameworks applied to national disease surveillance data used by WHO headquarters and government decision-makers, and trained over 500 staff in structured evaluation methods. Defining what "correct" means, then defending that definition to people who will challenge it, is the job.

Information systems that have to stay up when the stakes are real. Twelve years with the World Health Organization building and maintaining the Integrated Disease Information Management System and Pakistan's national Emergency Operations Centre dashboard, including the web APIs connecting them to HR, resource-planning and mobile data-collection systems.

That work was integration under pressure: multi-step data flows between systems that did not agree on formats, owned end to end, running through outbreak response on public-sector budgets. We bring the same approach to commercial clients — architecture, APIs, databases, dashboards, and the documentation that lets someone else maintain it.

Our point-of-sale and retail management platform, built on Laravel and PostgreSQL. It is in live daily use at a Pakistani retailer, where it replaced a commercial product — real transactions, real stock, real month-end.

Hisabifi covers sales, inventory, suppliers, receipts and reporting for independent retailers who have outgrown a cash drawer and a spreadsheet but are poorly served by enterprise systems. Available now for pilot deployments.

A CRM built for digital marketing agencies, live and in use. It pulls leads in from social media campaigns, puts them in one pipeline instead of three inboxes, and keeps call handling and follow-up attached to the lead rather than to whoever happened to answer.

Agencies lose revenue in the gap between a campaign generating interest and someone actually calling back. LeadFlow closes that gap: every enquiry lands in one queue, each one carries its source and history, and nothing waits on a spreadsheet being updated.