Vibhavari BellutagiRésumé
Vibhavari Bellutagi

Open to senior data & full-stack roles

I build the data platforms and the products that run on them.

Software engineer, 10+ years. Seven of them on data platforms at Deloitte, AgileLab and Mews. The hard parts get written down here. 19 posts so far.

Start learningAbout me

1 / 6

#apache-spark

Two DataFrames with 5 and 6 partitions, one join. Why does the Spark UI show 200 tasks?

Previously at
  • Mews
  • AgileLab
  • Deloitte

01 / Learn

Pick a topic. Leave knowing how it works.

Every post starts from something that surprised me, then explains the mechanism with a small worked example.

All 19 posts →

Read in this order0 of 4 read

What Spark is doing while you wait: nulls, jobs, stages, tasks and where the driver runs.

  1. Handling Nulls in SparkSix of ten movies have no rating, and filtering for == None returns zero rows with no error. How nulls behave in Spark, and the functions that handle them properly.Jan 13, 2025 · 8 min read
  2. Under the hood of a Spark jobTwo DataFrames with 5 and 6 partitions, one join, and the Spark UI shows 200 tasks. Where that number comes from, and how Spark splits work into jobs, stages and tasks.Jan 21, 2025 · 7 min read
  3. Spark Execution ModesSubmit a Spark job from your laptop, close the lid, and the job dies. Change one flag and it keeps running. Cluster, client and local mode come down to where the driver runs.Feb 6, 2025 · 6 min read
  4. The Life Cycle of a Spark Application ( Outside )You run spark-submit and, a while later, get a result. In between, four things happen and none of them are in your code: the request, the launch, the execution and the shutdown.Feb 7, 2025 · 4 min read

02 / Side projects

Each app has one hard part. That is the lesson.

Real apps with real constraints: payments, concurrency, auth. The hard parts become blog posts.

The hard part

Two members, one spot left.

Gym & studio management platform

Class scheduling with session credits, attendance, invoicing and online payments, with separate admin, coach and member portals.

TypeScript · Next.js · tRPC · PostgreSQL · Better Auth

How Postgres row locks prevent double bookings →

The hard part

Keeping a rent ledger correct when payments arrive partial, late or twice.

PG & hostel management platform

Multi-property dashboard, floor-room-bed layouts, a rent ledger that tracks paid, partial and overdue automatically, PDF invoices and role-based staff access.

TypeScript · Next.js · tRPC · PostgreSQL · Clerk

Post coming

03 / About

Who is writing this.

Portrait of Vibhavari Bellutagi
pipeline runtime
10h+ → 2h
compute cost
−60–75%
model deploy time
−50%
profile update latency
−75–80%
workflow efficiency
+40%

Software engineer, 10+ years. Seven of them spent on data platforms (Spark, Kafka, Databricks, Terraform) at Deloitte, AgileLab and Mews. On the side, I build full-stack products end to end, from the data model to the UI.

  • Data platforms

    Batch and streaming pipelines, CDC, lakehouse architecture and the infrastructure-as-code that keeps it all reproducible and cheap to run.

    Databricks · Spark · Kafka · FlinkSQL · Terraform

  • Backend & APIs

    Type-safe APIs, auth and payments, built on Postgres schemas designed to hold up under real concurrency.

    Python · FastAPI · TypeScript · tRPC · PostgreSQL

  • Product engineering

    Full products, from data model to UI to deployment, with monitoring, backups and the boring parts done properly.

    Next.js · React · AWS · GCP · Railway

Download résumé

Where I have shipped.

  1. Jul 2025 – Aug 2026France

    Data Engineer, Platform · Mews

    • Redesigned the platform architecture around configuration-driven infrastructure-as-code with Terraform on Databricks.
    • Built a batch CDC pipeline moving ~100GB from Postgres into Databricks.
    • Cut critical pipeline runtime from 10+ hours to 2 hours and compute costs by 60–75%.

  2. 2021 – 2024France

    Data Engineer · AgileLab

    • Built real-time pipelines with FlinkSQL and Kafka CDC, reducing profile update times by 75–80%.
    • Deployed NLP models on AWS with data scientists, cutting deployment time by 50%.
    • Delivered GDPR-compliant anonymization with sub-3-second query performance.

  3. 2018 – 2021Bengaluru

    Big Data Engineer · Deloitte

    • Built and optimized large-scale ETL on Hadoop, Spark and Hive, processing 30GB+ daily.
    • Created orchestration and logging frameworks that improved workflow efficiency by 40%.
“A real sense of ownership. She follows through, keeps herself accountable, and does not need to be pushed to deliver high-quality work.”
Elliot Ip · Senior Data Engineer, Mews
“Highly autonomous, proactive, and has a remarkable long-term vision.”
Vincenzo Pafundi · Machine Learning Engineer, AXA Italia
“In just a few weeks, she mastered the technical skills to work independently and soon took full ownership of entire projects.”
Marco Vagaggini · Senior Software Engineer, QuestIT
“She brought several ML projects to production on the AWS platform.”
Paolo Cristofanelli · ML & Data Platform Engineer, Cleo
Read all 6 recommendations on LinkedIn →

04 / Contact

Hiring for a data or full-stack role?

Senior data engineering, platform or full-stack product roles. Remote or hybrid. Questions about a post are welcome too.

Download résuméLinkedInEmail