Most Power BI dashboards die on arrival. After building 40+ dashboards for aluminium market intelligence at AlCircle, I found three patterns that separate dashboards people actually use from the ones they screenshot once and forget.
How I built an automated aluminium market intelligence system at AlCircle that ingests 500+ data sources daily, detects price anomalies within 2 hours, and gives our team a 3-day lead on market shifts that competitors report on a week later.
When I joined Mechanismic as a data intern, the team spent 6 hours every Monday manually pulling data from 500+ sources. I built a Python pipeline that reduced it to 12 minutes. Here's the architecture, the gotchas, and the code that made it happen.
Introducing the research feed — where data science meets industry insights. Follow along as I explore market trends, build dashboards, and push the boundaries of data-driven analysis.
A truncated y-axis, a pie chart with 12 slices, dual axes that imply correlation where none exists — I found all three in a single executive dashboard last quarter. Here's how to spot visual lies in your own reports before your stakeholders make bad decisions.
Everyone can write a SELECT statement. The analysts who get promoted write window functions in their sleep, use CTEs for readability, and know exactly when a subquery will kill performance. Here are the 5 patterns that changed my career.
BERT is great at understanding context but chokes on documents longer than 512 tokens. Mamba handles long sequences efficiently but lacks BERT's nuanced language understanding. For my M.Sc. research, I combined both — here's what worked, what didn't, and when you should just use a simpler approach.