Reading this as a hiring manager for
Sofia MarchettiAnalytics engineer
I build the tested data models that dashboards and models quietly depend on.
PythondbtAirflowSnowflakeKafkaTerraform
Six years in data at a grocery chain, a freight company and a digital bank. I have built the dashboards, the pipelines under them and the models on top, so I can start wherever your team needs the most help.
One person, four job titles
I started as the analyst who rebuilt a grocery chain's weekly margin report, moved into forecasting delivery delays for a freight company and now run the data platform and fraud models at a digital bank. Each move came from following the work one layer deeper.
That path means I can sit in any of these four seats. I would rather tell you which one fits me best for your team than pretend to be a specialist in all of them, so pick the role you are hiring for and the page reorders itself around it.
- Based in
- Milan, Italy
- Experience
- Six years, three teams
- Works with
- Product, finance and operations teams
- Available
- From January 2027, hybrid or remote
Strengths
The same habits show up in every role. What changes is where I spend most of the week.
Pipelines that fail loudly
Freshness checks, row count tests and schema contracts on every source, so a broken feed stops the run and pages someone instead of filling a dashboard with yesterday's numbers.
Data modeling for the next reader
Staging, intermediate and mart layers with consistent names and documented grain. A new analyst can find the right table in minutes, not days.
Traceable runtime costs
Every model tagged by owner and use case, with warehouse usage reported weekly. At Ardesia Bank that cut nightly build time by 40%.
Calm backfills
Incremental models designed so a three year backfill is a planned job with a runbook, not a weekend spent watching queries.
Looking for a team that treats its warehouse like a product: code review, tests in CI and owners for every model.
Toolkit
One stack, used differently depending on the job. The tools for the selected role stay highlighted.
Query
- SQL
- Python
- R
Data modeling
- dbt
- Snowflake
- Airflow
- Kafka
- Terraform
Delivery
- Looker
- Streamlit
- FastAPI
- Docker
Statistics and ML
- scikit-learn
- statsmodels
- PyTorch
- MLflow
Selected work
Sorted for platform work: the ingestion, the model layer and what it takes to run.
Shared metric layer
Ardesia Bank, 2024
Finance, risk and product each reported a different number of active customers. I moved 60 core metrics into dbt semantic models with written definitions and a named owner for each.
Analytics engineer: Designed the mart layer, added contract tests on every source and cut the nightly build from 90 to 54 minutes.
340 overlapping dashboards cut to 58 and one active customer figure in the board pack.
Parcel scan event pipeline
Velaria Freight, 2023
Depot scans landed in the warehouse once a night, so dispatch planned each morning on data that was already a day old. I moved the scan feed onto Kafka with streaming loads into Snowflake and tested dbt models on top.
Analytics engineer: Owned the ingestion, the Airflow schedules and the Terraform for every environment.
Data freshness went from 24 hours to under 15 minutes. Pipeline incidents fell from nine a quarter to two.
Store margin reporting
Pianura Grocers, 2021
Weekly margin for 96 stores lived in a chain of spreadsheets that took two people most of Monday to refresh. I rebuilt it as dbt models with a Looker dashboard on top.
Analytics engineer: Wrote the first dbt project at the company, with tests that caught two broken supplier feeds in the first month.
The Monday report now refreshes itself by 7 am and is opened weekly by 85 of 96 store managers.
Real time card fraud scoring
Ardesia Bank, 2025
Fraud rules flagged so many good payments that customers were calling in to unblock their cards. I replaced the rules with a gradient boosted model behind a FastAPI service, fed by a feature store of spending patterns.
Not the closest match for this role, kept here for the full picture.
22% more fraud caught at the same false alarm rate, with blocked good payments down by a third.
Next day delivery delay forecast
Velaria Freight, 2023
Late deliveries were spotted only after customers complained. I built a model that scores every route the evening before, using depot load, driver hours, weather and the route's own history.
Not the closest match for this role, kept here for the full picture.
On time delivery rose from 88% to 93% across 14 depots in the first six months.
Promotion uplift testing
Pianura Grocers, 2021
Promotions were judged on raw sales, so every one looked like a success. I set up matched holdout stores and measured the real lift against what each store would have sold anyway.
Not the closest match for this role, kept here for the full picture.
Four of eleven recurring promotions were losing margin and were dropped the next quarter.
Experience
2024 to now
Senior analytics engineer, Ardesia Bank
Milan, Italy
Lead the data platform for a digital bank with over a million customers. Own the dbt project, the shared metric layer and the fraud scoring service, working with a team of four engineers.
2022 to 2024
Data scientist, Velaria Freight
Milan, Italy
Forecasting and operations research for a parcel network of 14 depots. Built the delivery delay model and rebuilt the scan event pipeline it depended on.
2020 to 2022
Data analyst, Pianura Grocers
Bergamo, Italy
Commercial analytics for a regional grocery chain. Owned weekly margin reporting and promotion analysis for category and store teams.
Education
2018 to 2020
MS in Data Science
Lombardy Institute of Technology
Thesis on demand forecasting for perishable goods with sparse sales history.
2015 to 2018
BS in Statistics
Northern Italy State University
Focus on probability, survey sampling and applied regression.