MBILAL.WORKS / V3 · RESEARCHVOL. I · ISSUE 01 · APR 2026
FIELD NOTE №1● OPEN ACCESS4-MIN READ

Field notes from shipping applied AI in production.

BY MUHAMMAD BILAL·GERMANY·CORRESPONDENCE: mbilal.workmail@gmail.com

ABSTRACT —Muhammad Bilal is an applied AI consultant based in Germany who builds agentic AI systems that deliver measurable outcomes for clients — from strategy and architecture through to production delivery. This site collects three engagements — a BI stack, an agentic tool, and a research pipeline — and the lessons that generalise across them.

KEYWORDS:applied-AIagentic-systemsPower-BILLM-pipelinesclimate-disclosureGermany

Replacing a €50,000/yr controlling role with a BI stack and 23 KPIs.

A mid-sized German marketing agency (~20 employees) lacked end-to-end visibility into billable efficiency across project and service teams. Over ten weeks I designed and deployed 8 Power BI dashboards backed by a Node/Express middleware bridging Absence.io (HAWK auth) and Toggl (API-key auth) — neither of which Power BI supports natively. First month of deployment moved project-team billable efficiency from 49% to 65% (Δ +16 pp). The COO attributed €50,000/yr in controlling-role cost savings to the new reporting surface.

TABLE I — RESULTS
Cost savings (COO-attributed)€50,000 / yrfull-time controlling role
Project team efficiency49% → 65%first month of deployment
Dashboards shipped8production
KPIs modelled23cross-team
PoC → contract186 hrsphase 2 in progress
Read the engagement

Meeting notes to a fileable Jira backlog in five minutes.

Planning meetings generate messy transcripts. Jira expects clean epics, stories, acceptance criteria, and estimates. This open-source tool bridges the two: it parses raw notes, decomposes them into an epic-and-story tree with acceptance criteria, and files the result against the Atlassian REST API. Built for teams tired of the Monday-morning ticketing tax.

TABLE I — RESULTS
Notes → filed backlog≈ 5 minper sprint
Manual equivalent~ 60 minsprint planning
Output ticketsEpic + Stories + ACstructured
RuntimePython · PydanticCLI / action-hook
See it in action

An LLM extraction pipeline for STOXX Europe 600 climate-disclosure research.

A TRR 266 working paper on climate disclosure required structured extraction of financial-statement paragraphs across six years of STOXX Europe 600 annual and audit reports — at a scale where manual coding was untenable. I built the scraping and extraction pipeline: PDF ingestion, LLM-backed section identification, structured output for downstream NLP with ClimateBERT. The working paper, co-authored by researchers at LMU Munich, Bocconi, and IESE, cites the work in its acknowledgements.

TABLE I — RESULTS
Firms covered600STOXX Europe 600
Reporting years62018–2023
Downstream corpus6.3M+ paragraphsfinancial statements
AcknowledgementSSRN 4763140Müller et al., 2024
Read the paper
§04
CAPABILITIES

Four practices, one shared criterion: does it reach production?

№PRACTICEDESCRIPTIONREPRESENTATIVE WORKSTACK
01Agentic SystemsMulti-step LLM pipelines with tool use, retrieval, and evaluation. Built for production latency and cost constraints.Jira-Ticket-Generator, CV-TailorClaude · Pydantic · DSPy · Atlassian
02BI & AnalyticsKPI architecture, Power BI modelling, Node middleware to bridge auth protocols across vendor APIs.LaMechKy BI stackPower BI · Node · Express · Render
03Research EngineeringData infrastructure for academic and regulatory research — PDF ingestion, LLM extraction, reproducible pipelines.STOXX 600 scraper (SSRN 4763140)Python · Claude · Pandas · DuckDB
04Ops AutomationInternal tools that collapse hours of operations work — note-to-ticket, CV-tailoring, report generation.Granola-to-Notion, Professional-Profile-MakerClaude · Notion · Scriptable · Python
§05
CORRESPONDENCE

For collaboration, consulting, or comment:

I take on a small number of applied AI engagements per quarter. Send a short note with the problem, constraints, and timeline — I reply within 48 hours with an honest read on whether I can help.

TYPICAL RESPONSE
≤ 48 hours
ENGAGEMENT SHAPE
6–14 weeks, fixed-scope
TIME ZONE
CET (UTC+1 / +2)