Abdullah Akbar Khalid

Companies House Screening Engine[1]

Which founder-owned tech-services firms actually fit a roll-up thesis, including the ones no keyword search would ever surface? This engine takes the entire UK register (~5M companies, 400+ columns) down to a ranked, defensible shortlist: it discovers each company's website, classifies it with rules + Claude on a four-axis taxonomy, recovers financials from iXBRL accounts, and scores defensibility.

Covenant[2]

A lease is offered by entity X. Is that covenant real? Search the register for "Pret A Manger" and twenty entities come back, two of them named almost identically. One is the trading company filing full accounts since 1984. The other is a 2018 acquisition vehicle, formerly JAB (ACQUISITION) LTD. Same brand, same registered address, opposite covenant.[3]

Covenant screens tenant strength from free, public Companies House data (one tenant, a rent roll, or the whole country) instead of a per-report credit agency fee.

Oversight[4]

Since 9 March 2026, a RICS member whose AI use materially affects surveying work has to be able to show four things: which AI systems the firm uses, that a named qualified surveyor wrote a decision on how reliable the output was, that the client was told in the Terms of Engagement, and that the vendor was diligenced.[5] That is a records problem, and I could not find a single tool built for it.

Oversight keeps those records as plain files in a git repository, so authorship, timestamps and append-only history come free: the audit trail is the repo. check exits non-zero and names the obligation each failure would breach. pack renders the regulator-ready evidence pack and refuses to run while any check fails, so a pack existing is itself evidence the records were coherent when it was made.

The example firm registers Covenant as one of its AI systems, with a real reliability decision record for a real screening. One tool doing surveying work with AI, one tool governing AI to surveying standards.

cited-memo[6]

AI drafts a plausible investment memo in seconds. The problem is the next hour, when an associate re-checks every figure against the data room because nobody can tell which numbers came from the accounts and which the model invented. cited-memo inverts it: it drafts from source documents only, cites the exact page behind every figure, then grades its own output (offline, no API key).

You check the two flagged sentences instead of all twenty-six.

Command Center[7]

One dashboard for what six apps and my head used to hold: today's priorities, the backlog, overdue items, people I owe a reply, the job pipeline. My AI automations capture into it all day. The page that justifies the project is Automation Health, because automations fail silently and a briefing that stops arriving looks identical to a quiet news day until something checks.

The AI-Impact Map[8]

"Which industries will AI eat" is usually answered at the level of a category, which is the wrong altitude: AI acts on tasks, and every category is a bundle of tasks with wildly different exposure. So I scored 681 UK business-model subsectors individually on nine anchored criteria, across two independent axes, and attached live Companies House counts so the size of each bubble is a real number of real companies.

The finding that mattered was not the answer. Scoring 50 subsectors twice with two different models as judges produced only 33% exact agreement, and the disagreement was systematic rather than noisy. When your quadrant boundary is a median, mixing judges silently sorts one judge's rows into different quadrants than the other's. I re-scored all 681 with one judge.

What a job board says that a strategy deck does not

A strategy deck is a claim. An open requisition is a budget somebody already signed. So I read the second one: point this at any company's public job board and it returns the investment mix by region and function, ready to set against where the revenue actually comes from.

Run against Elastic's 250 live requisitions, the gap is immediate. International is 46% of revenue but 63% of open roles. The mix, though, inverts: outside the US more than half of hiring is engineering, while inside the US nearly half is field. The US alone holds more field requisitions than the whole of EMEA. Read together, a good share of what looks like international expansion is engineering capacity placed in lower-cost hubs, and the question that follows is where International should add field capacity next.

See it used in a market prioritisation model, which uses this tool for the coverage and competitor halves of the analysis.

The systems that run my day

No repo for these; they hold my sessions and keys. A scraping engine that reads pages bot-detection blocks, including LinkedIn. A Discord to WhatsApp relay that reformats a trading community's calls into structured tables in real time. An Instagram story archiver with review notes. And a briefing pipeline that ranks everything I'm tracking and pushes a top-five to my WhatsApp twice a day, on a launchd schedule.