siva.Let’s talk
DATA ENGINEER / AI EXPLORERSELECTED WORK · 2026

Reliable data.
Measurable AI.

I’m Siva. I build data platforms that hold up.
And explore AI systems you can actually evaluate.

Explore selected work GitHub
Siva
Currently at ChargebeeData platforms · Chennai, India
SYSTEMS, WITH INTENTION[ 01 / 03 ]
RAW INPUTSTRUCTUREPOSSIBILITY
COMPLEXITY IN.
CLARITY OUT.

Built around one question: can you trust the result?

FROM SOURCE TO SYSTEM
IN THE REAL WORLD

Engineering at scale.
Impact in the details.

From my LinkedIn
3.2+TB / DAY

Data platform throughput

83%LESS LATENCY

API performance improvement

64%LESS RUNTIME

Batch processing improvement

01 / SELECTED ENGINEERING

Small systems.
Serious questions.

Six recent projects across reliable data and measurable AI. Open the notes, inspect the trade-offs, run the code.

All repositories
06 PROJECTS
01 / INGESTION & MODELING

Messy payment events. A trustworthy revenue mart. A pipeline you can replay without counting twice.

PythonSQLiteIdempotency
Source
02 / RETRIEVAL & EVIDENCE

Before generating an answer, inspect the evidence. BM25 retrieval with citations and repeatable evaluation.

PythonBM25Recall@k / MRR
Source
03 / STATE & CONSISTENCY

Late updates, duplicate deliveries, and deletes. Rebuild entity state without bringing old data back to life.

PythonSQLiteTransactions
Source
04 / EVALUATION & REGRESSION

Make saved AI responses testable: structured output, citation checks, abstention, coverage, and a CI gate.

PythonJSON contractsEvaluation
Source
05 / QUALITY & CONTRACTS

Catch drift at the boundary. Explicit CSV contracts with freshness, uniqueness, and actionable findings.

PythonData qualityCLI
Source
06 / DATA FOR MACHINE LEARNING

Train on what was actually known. Historical features that respect both event time and availability time.

PythonTemporal SQLML features
Source

Independent, AI-assisted learning and reference projects. Synthetic fixtures, documented limitations, and reproducible demos—not production benchmarks.

THE EARLIER CHAPTERSFrom first experiments to better questions.
Transformer from scratchTensorFlow · English → FrenchBanking ETLPython · extraction → SQLBook StoreReact · Express · MongoDBFlutter WeatherFlutter · Dart · OpenWeather
02 / THE WORK BEHIND THE WORK

Built in
the real world.

From financial data platforms to subscription infrastructure. Making complex systems easier to trust—and easier to operate.

Full experience on LinkedIn
THE THROUGH LINE

Clarity in the design.
Reliability in the details.

Jul 2026 — PresentCURRENT

Chargebee

Data Engineer

Building and operating data workflows for billing, revenue, and subscriptions.

  • Consolidated Spark / Amazon EMR orchestration into a shared Python module and refactored Airflow scheduling with TaskGroups.
  • Built GitHub Actions deployment workflows and Datadog dashboards for export duration, volume, regional trends, and troubleshooting.
AirflowSpark / EMRPythonDatadogGitHub Actions
Oct 2024 — Jul 2026

KFin Technologies

Software Engineer

Financial data platforms, from on-premises sources to curated lakehouse datasets.

  • Delivered data architecture for a mutual fund platform serving 11 million users and a network of 1 million distributors.
  • Built full and incremental Apache Iceberg ingestion and a data lake API buffering approach that loaded 2 TB into PostgreSQL in 30 minutes.
Apache IcebergAWSPostgreSQLData modeling
Apr 2024 — Oct 2024

KFin Technologies

Software Engineer Intern

Cloud ELT and reliable delivery for financial reporting and investor analytics.

  • Built Snowflake pipelines from APIs, S3, and SQL Server using Snowpipe, Spark, and Airflow.
  • Reduced manual reconciliation effort by 70% and improved reporting and analytics performance by 45% through ingestion and workload optimization.
SnowflakeSnowpipeSparkSQL Server
BEFORE THAT

Assistant Software Developer · Amaravathi Software Innovations
Jul 2023 — Apr 2024

Student Teaching Assistant · Bennett University
Jul 2022 — Jan 2023 · Mentored nearly 40 students

03 / TRY THE FAILURE PATH

The second run
is the real test.

A good pipeline works once.
A reliable one knows what to do next.
Replay a batch. Inject a failure. Inspect the result.

RELIABILITY LABLOCAL DEMO / SYNTHETIC DATA
PAYMENT EVENT STREAMJSONL → LEDGER → MART
01pay_001paymentUSD12,000
02pay_002paymentUSD8,000
03ref_001refundUSD2,000
04pay_003paymentINR50,000
05pay_002paymentUSD8,000repeated
06bad_001paymentUSD-500invalid

Amounts are integer minor units. Refunds subtract from their own currency.

INJECT A FAILURE
REVENUE MARTRUN 00
USD / MINOR UNITS0
INR / MINOR UNITS0
0 unique events in the ledgerCURRENCIES ISOLATED
Ready to inspect

Start with six sample events. Then try to break the assumptions.

↳

Results appear here.
Every event gets an explanation.

A small JavaScript model of replay, validation, and currency isolation. The Python repository adds raw storage, timestamps, transactions, and SQL.

Inspect the full implementation
04 / BEHIND THE SYSTEMSA LITTLE ABOUT ME
Padala Bala Siva Sai Megi Reddy
SIVA / CHENNAI, INDIA

Curiosity starts it.
Craft carries it.

I like the moment when a complicated problem starts to make sense.

My work lives between data engineering, backend systems, and practical AI. I care about the edges: the event that arrives twice, the data that arrives late, the answer that looks right but has no evidence.

That curiosity started with computer science at Bennett University and grew through teaching, full-stack development, and financial data platforms. Today, I bring it to my work at Chargebee and the projects I build outside it.

BENNETT UNIVERSITY · 2020—2024B.Tech. in Computer Science
9.21/ 10 CGPA
TOOLS I WORK WITH

Python · SQL · Spark · Airflow · AWS · Snowflake · Iceberg

The tool changes. The engineering principles stay.
GOOD WORK STARTS WITH A CONVERSATIONCHENNAI / --:-- IST

Let’s build
something that lasts.

A data problem. An ambitious idea. A different perspective.
I’d love to hear what you’re thinking.

GitHubLinkedIn