Monil
Soni
- Power BI Developer
- Business Intelligence Analyst
- Data Analyst
Over five years turning scattered business data into reporting people actually rely on. I model it properly, write the DAX and SQL underneath, and own the Power BI Service side — so the report still works the morning after it's handed over.
Currently at Smart Minds · 30-day notice, negotiable · open to relocation & remote
It really cross-filters — click a role. Unselected bars dim rather than
disappear, because removing context is how you lose the reader.
Projects
Problems someone needed solved.
Each one starts with the question that was actually asked, not the tool I opened. Below the line, GitHub feeds itself in — push a repo and it appears here.
Why does the refresh keep failing overnight?
Re-architected enterprise semantic models into star schema and refactored the worst DAX and M using DAX Studio, Tabular Editor and VertiPaq Analyzer.
How do 200 people see only their own region?
Designed RLS roles mapping region, business unit and hierarchy onto a single certified dataset — instead of cloning reports per team, which is how governance dies.
Can we stop rebuilding this in Excel every Monday?
Migrated legacy VBA and Excel reporting into automated Power BI datasets, then standardised KPI definitions and SOPs across regional units.
Case study
Cutting refresh time by 55% without touching a visual.
The engagement I'd most want to be asked about. Read it in order — it's the order an interviewer thinks in.
- Where
- Smart Minds, 2024—present
- Role
- Sole owner — modelling, DAX, Service
- Audience
- 200+ users across regions and business units
- Tooling
- DAX Studio, Tabular Editor, VertiPaq Analyzer, Best Practice Analyzer
- Result
- −55% refresh, 99% accuracy, −40% ad-hoc requests
The question
Refreshes were failing, and nobody trusted the numbers by Friday.
Scheduled refresh was overrunning its window and dropping out. Every failure meant a manual re-run, and every re-run meant someone in ops quietly went back to their own spreadsheet. The trust problem was downstream of the performance problem.
What I found
A flat model doing work the storage engine should have done.
VertiPaq Analyzer showed the damage: wide flat tables with high-cardinality text columns eating memory, relationships pointing in both directions, and measures written as row-by-row iterators where a simple aggregation would do. Power Query was breaking query folding early, so filtering happened in the mashup engine instead of at the source.
Nothing here was a visual problem. It was a modelling problem wearing a visual problem's symptoms.
What I changed
Star schema, single-direction relationships, folding restored.
Split the flat tables into a fact table with narrow conformed dimensions and a marked date table so time intelligence stopped guessing. Removed bidirectional cross-filtering except where genuinely needed. Reordered M steps so filters happen before anything that breaks folding, and pushed the heavy joins back into T-SQL where they belong. Best Practice Analyzer caught the rest.
-- iterator replaced with a filtered aggregation; same result, fraction of the cost
Sales YoY % =
VAR Curr = SUM( Fact_Sales[NetAmount] )
VAR Prior = CALCULATE( Curr, DATEADD( Dim_Date[Date], -1, YEAR ) )
RETURN DIVIDE( Curr - Prior, Prior )
The governance layer
One dataset, not forty copies.
The easy answer to "each region needs its own view" is a report per region. That's forty artefacts drifting apart within a quarter. Instead: RLS roles mapping region, business unit and hierarchy onto the single certified dataset, so 200+ users get self-service without anyone getting a second version of the truth.
What changed
55% faster, no more overnight failures, 40% fewer ad-hoc requests.
Reported accuracy reached 99% once the M validation routines were in, and the ad-hoc queue dropped by 40% because people could answer their own questions. The thing I'd do earlier next time: run Best Practice Analyzer on day one rather than week three — it flagged issues I'd already spent two days finding by hand.
Background
Experience & education.
Laid out as a tree, because that's how the work nests — each role branches into what it produced.
-
Smart MindsPower BI DeveloperNov 2024 — Present
- Re-architected enterprise semantic models into star schema and refactored DAX and M with DAX Studio and Tabular Editor — refresh time −55%, scheduled failures eliminated.
- Designed RLS roles across region, business unit and hierarchy on one governed dataset — 200+ users with secure self-service, replacing 40 duplicated reports.
- Own Power BI Service end to end: workspace and app lifecycle, dataflows, on-premises gateway, refresh scheduling — the manual weekly status reporting cycle retired.
- Build M ETL pipelines and DAX measures with time intelligence and what-if parameters, plus reconciliation checks validating published figures against source systems.
- Deliver Tableau and Looker Studio views alongside the certified Power BI layer for teams already working in those tools.
-
PW (PhysicsWallah)Power BI DeveloperMar 2023 — Oct 2024
- Built interactive Power BI and Tableau dashboards mapping critical business pipelines — turnaround −50% for business and operations teams.
- Migrated legacy VBA and Excel reports into automated Power BI datasets — 20 hrs/week saved, and one certified source of numbers.
- Modelled source data into star schema and tuned T-SQL against high-volume SQL Server tables to feed certified datasets.
- Standardised data definitions, KPI logic and SOPs across regional units — discrepancies −35%.
-
United Cable NetworkData AnalystAug 2019 — Mar 2021
- Supported data and business planning for the commercial team, owning BI and analytics requests from the question asked to the report delivered.
- Automated and simplified recurring reporting trackers using Advanced Excel, Power BI and SQL — repeated manual consolidation removed.
- Created and maintained the commercial data catalog, standardising definitions and sources across teams.
- Ran periodic data quality reviews across internal and external data, resolving gaps before they reached reporting.
-
MBA, Business AnalyticsParul University, Vadodara2021 — 2023
-
BBA, Information Systems ManagementSardar Patel University, Vallabh Vidyanagar2016 — 2019
-
Certifications
- Business Intelligence Certification — Rishabh Software 2022
- Corporate Readiness & Adaptability Program — BIT Baroda
What I actually do
Two halves of one job.
Plenty of people can build a report. Fewer can build the model underneath it and then run it in production. The second half is where the 55% came from.
01 — Model
Modelling & performance
- Semantic design — star and snowflake schema, conformed dimensions, composite models, aggregations, marked date tables
- DAX — time intelligence, calculation groups, dynamic measures, what-if parameters, context transition used deliberately
- Power Query (M) — custom functions, validation routines, and query folding that survives the whole chain
- Tuning — DAX Studio, VertiPaq Analyzer, Best Practice Analyzer, storage-engine vs formula-engine diagnosis
- SQL — T-SQL on SQL Server and Azure SQL: CTEs, window functions, indexing, execution plans, stored procedures
02 — Run
Service & delivery
- Governance — Row-Level Security, certified datasets, deployment pipelines, Premium/PPU capacity, usage monitoring
- Operations — workspace and app lifecycle, on-premises gateway, dataflows, incremental refresh, XMLA endpoint
- Definitions — KPI governance, data dictionaries, data catalogs, SOP standardisation across regional units
- Stakeholders — requirements workshops, UAT, root-cause analysis, executive reporting
- Report craft — layout that reads in ten seconds, colour reserved for meaning, paginated reports where they fit
Toolbox
The honest version.
Split by how often I actually reach for it, not by logo count.
Next
Let's talk about
your model.
Currently with Smart Minds on a 30-day notice period — negotiable for the right role — and open to relocation and remote. Email is fastest, I reply within a day. Happy to walk through any model, measure or RLS design above.