moji.dev AI · DATA · MILANO

moji.dev ── system bootstrap

phase init
identity-verify moji@milano · OK
tls-handshake ECDHE-AES256 · 1.3
vault-unseal kv://milano-001
azure-data-factory 1 847 392 rows/s
databricks-cluster 42 active jobs
synapse-pool DW400 · 12% idle
sql-server-primary v2025 · 99.97% up
neo4j-graph 2 134 982 nodes
kafka-stream topic: events.v3
openai-api gpt-4o · 420ms p50
azure-ai-foundry vector store · 14ms
foundry-local phi-3 · 13.7ms/tok
embedding-cache 847 392 vecs
eval-runner golden_set/v7 · 0.847
guardrails-v2 12 policies
drift-monitor 0.012 < 0.05
cost-tracker $0.0042/req · -87%
ntfy-channel subscribed
[boot] identity verified · moji@milano
[boot] tenant moji.dev · milano-001
[boot] tls ECDHE-AES256-GCM-SHA384
[adf] pipeline bronze_ingest · green · 4/4
[databricks] cluster warming · 4× DS4_v2
[databricks] delta OPTIMIZE · 17.4M rows
[synapse] dw resumed · queue: 0
[neo4j] query MATCH path · 62ms
[kafka] consumer ai-svc-prod · lag: 0
[openai] warming gpt-4o · 32k ctx
[foundry] vector upsert 1024 · cosine 0.91
[foundry-local] load phi-3-mini · 1.8 GB
[eval] golden_set/v7 0.847 · 0.812
[guardrails] jailbreak blocked · ptn 17
[drift] kl_div 0.008 · 0.012 OK
[finops] cost $0.0042 · savings 87%
[ntfy] backpressure 78%
[boot] SYSTEM READY
streaming sources…
AI INTEGRATION · EDGE · DATAOPS · MILANO 🇮🇹
DATA SCIENCE · APPLIED ML · MILANO 🇮🇹
DATA ENGINEERING · AZURE · MILANO 🇮🇹
BI · POWER BI · DAX · MILANO 🇮🇹
DATA ARCHITECTURE · LAKEHOUSE · MILANO 🇮🇹

From demo to production.
AI that runs.

Models that ship,
not just notebooks.

Pipelines that don't
wake you at 3 a.m.

Warehouses
turned into decisions.

Platforms for the
boring quarter.

LangChain · PromptFlowAI AgentsAzure AI FoundryFoundry Local · EdgeDatabricks · ADFMLOps · Evals
Python · scikit-learnXGBoost · LightGBMPyTorch · HuggingFacestatsmodelsSHAP · causalMLflow · evals
Azure Data FactoryDatabricks · SparkSynapse · Fabricdbt · AirflowKafka · Event HubsTerraform · CI/CD
Power BI · DAXTabular EditorVertiPaq · DAX StudioSSAS · semanticSQL Server · SynapseSSIS · ADF
Azure architectureLakehouse · MedallionKimball · Data VaultPurview · Unity CatalogTerraform · BicepFinOps

About

I take AI from notebook demos to systems that run.

The hard part of applied AI isn't the model. It's the integration: stable orchestration, eval gates that fire before users notice, guardrails that hold, latency budgets that don't blow up the finance review.

Data Scientist who ships models, not notebooks.

Boring stack first — gradient boosting, classical NLP, embeddings. Deep learning only when it's the cheapest answer. Forecasts come with calibrated confidence intervals.

Pipelines that don't wake you at 3 a.m.

Bronze/silver/gold on Databricks + ADF with proper SLAs; validation, secret scanning, trigger gates on every PR; dbt with tests and lineage; IaC from day one.

Warehouses turned into decisions.

The hardest part of BI is the semantic model. Power BI semantic models executives can trust, DAX tuned with VertiPaq Analyzer, governance that scales.

I lead with: "who pages whom at 3 a.m.?"

End-to-end platform design on Azure with governance, lineage, FinOps baked in from day one — not bolted on in year two.

— "Thoughtful before fancy."

What I do — in detail

01 · DATA DESIGN

Models that survive contact with humans

Star and snowflake where they fit. Data vault when history matters. Contracts at every boundary so producers can refactor without breaking downstream.

Toolbox · Kimball · Data Vault 2.0 · dbt · Great Expectations
02 · ARCHITECTURE

Lakehouse, lambda, kappa — choosing what fits

Event-driven where it earns complexity. Batch where SLAs allow. Cost-aware always.

Toolbox · Azure · Terraform · Bicep · Kafka · Delta Lake
03 · SCIENCE & ML

Boring stack first, deep learning when it pays

Gradient boosting, classical NLP, well-engineered features. Foundation models go through eval gates before they touch users.

Toolbox · scikit-learn · XGBoost · PyTorch · LangChain · MLflow
04 · ANALYSIS & PREDICTIONS

Forecasts that respect the process

Seasonality, regime shifts, exogenous shocks. Calibrated confidence intervals. Causal methods when the question demands them.

Toolbox · statsmodels · pmdarima · EconML · SHAP
05 · GOVERNANCE

The work that decides payback or drift

Contracts at boundaries. Lineage that survives refactor. Drift detection. Cost dashboards someone outside engineering reads.

Toolbox · OpenLineage · Purview · Unity Catalog · FinOps

Experience — themes

Around a decade across data engineering and applied AI in Italy and the EU.

2024 — NOW

AI integration architect · freelance

Production LLM systems and AI agents on Azure. Edge AI on Foundry Local; CI/CD for ADF.

Azure AI Foundry · LangChain · Databricks · ADF
2022 — 2024

Senior data engineer · Azure data platforms

Lakehouse design — bronze/silver/gold on Databricks + ADF, dbt, CI/CD, cost optimization.

Databricks · ADF · Synapse · dbt · Power BI
2019 — 2022

Data & BI consultant · DsBiConsultant + freelance

Enterprise reporting in banking, insurance, media, retail. SQL Server modernization, Power BI semantic models.

SQL Server · Power BI · SSIS · DAX
2017 — 2020

Digital humanities · Università di Pisa

Network analysis, NLP on historical corpora, graph DBs for cultural heritage.

Neo4j · Python · NLP
2015 — 2017

Software & web · early career

Full-stack — PHP, JavaScript, Linux, MongoDB, early Docker.

PHP · JS · MongoDB · Docker

Stack

AI / ML
Azure AI Foundry · OpenAI · LangChain · PromptFlow · MLflow · Foundry Local · ONNX
Data
Azure Data Factory · Databricks · Synapse · Fabric · SQL Server · Spark · dbt · Neo4j
Platform
Azure · Terraform · Bicep · GitHub Actions · Docker
Languages
Python · PowerShell · TypeScript · SQL
Spoken
Italian · English · Persian

Featured work

Engage me for

Contact

Tell me about the problem. One-line briefs welcome. I reply within 24h.

Static site — manual send required. Submitting opens your email client with the message pre-filled. You'll need to click Send in your email app to deliver it. No data is sent automatically.
No newsletter · GDPR Art. 17.