I build AIthat ends uprunning.
And I have the receipts. A Power BI stack giving a twenty-person agency its first honest read on billable efficiency. An extraction pipeline rendering six years of STOXX Europe 600 annual reports as a cited research dataset. A ticket generator that makes sprint planning a five-minute task.
Read on↓Most AI work
never reaches
production.
95% of enterprises report zero measurable profit impact from AI adoption. The shortfall is not in the models. It is in the engineering, the operations, and the discipline required to make a model survive contact with a real organization.
I work in the gap between a working notebook and a system someone else can run on a Monday morning without calling me. That gap is where most projects quietly die — and where the real economics live.
Three that did.
Selected from sixteen — chosen because the receipts are countable, recent, and attributable. The rest live at /projects.
- i.Advisory engagement · Marketing & comms agency · NRW, Germany
LaMechKy GmbH
Ten weeks, settled as an unpaid PoC, turned into a paid engagement when the numbers held. A Power BI stack on top of Toggl, Absence.io, and a Node middleware bridging the auth protocols product teams pretend aren't there. Eight dashboards, twenty-three KPIs live by the close of phase I.
Power BI · Node · Express · Toggl · Absence.io · LaTool · Render
Read the engagement note→ - ii.Research pipeline · TRR 266 · TUM × LMU × Bocconi × IESE
STOXX Europe 600 extraction
Six reporting years of European annual reports rendered into a machine-readable corpus, with provenance back to the source page preserved. The downstream paper, on climate disclosure, runs the dataset through ClimateBERT. The acknowledgement is a small line of type. It has opened doors.
Python · Claude · PyMuPDF · ClimateBERT (downstream)
Read SSRN 4763140↗ - iii.Open source · Operations automation
Jira ticket generator
Planning meetings produce transcripts. Jira expects epics, stories, acceptance criteria, and estimates. Between the two lies a tedious hour. This tool collapses it to about five minutes — parsing, drafting, and filing against the Atlassian REST API in a way any team lead can audit.
Python · Claude · Pydantic · Atlassian REST · MIT-licensed
See the repository→
Small teams.
Fixed scopes.
Production targets.
- i.
Small teams.
I work directly with the person who can decide. No account managers, no project triangles.
- ii.
Fixed scopes.
Six to fourteen weeks, written down before we begin. No retainers, no scope drift, no surprise invoices.
- iii.
Production targets.
Every engagement closes on a measurable outcome — a number on a dashboard, a paper that cites the work, a system someone else now runs.
- iv.
Honest no.
If a project is wrong for me, or a model is wrong for the project, I will tell you in the first call. The reply is part of the work.
On the right
problem, I can save
you a year.
Send the problem, the constraints, the deadline. I reply within forty-eight hours and tell you honestly whether I am the right person for the work.
mbilal.workmail@gmail.com↗- Reply within 48 hours.
- Engagements typically 6 – 14 weeks, fixed scope.
- CET — UTC +1 / +2.