Building pipelines people can trust.
I build large-scale data platforms by day — streaming systems, lakehouses, enterprise RAG — and research explainable AI by night, making ML systems transparent enough to bet real decisions on. Scroll — the particles will follow.
Numbers that ship.
Things I've built.
Enterprise RAG for Financial Docs
A retrieval-augmented pipeline for financial document summarization — LangChain + vector databases wired into governed banking data flows, cutting manual review effort by 40%.
10x Lakehouse Migration
Moved RDBMS workloads to an Azure Databricks lakehouse — PySpark pipelines over 5M+ daily records running 10x faster, plus a 100GB+ Synapse warehouse consolidating 10+ vendor feeds.
ExplainOps — Open Source
An ops layer for ML explanations: lineage-linked attribution tracing, drift monitoring, and trust checks. Part of an XAI × Data Engineering series I'm shipping repo by repo on GitHub.
Trustworthy AI Research
Survey of Trustworthy & Explainable AI accepted at IEEE WorldSUAS 2026 — alongside PhD research on making LLM systems transparent, bias-aware, and auditable in enterprise data platforms.
Shipping in public.
One repo at a time, an XAI × Data Engineering stack — each with tests, CI and a runnable demo. Every claim below is something you can clone and check.
Explainability as a pipeline stage — versioned, partitioned, queryable explanations.
Trace a wrong prediction back to the corrupt upstream table that caused it.
Accuracy holds steady while the model's reasoning changes. Detect it without labels.
One trust report card for any classifier — fairness, calibration, robustness.
Do your feature-importance methods actually agree? Measure it, don't assume it.
Structured streaming with event-time windowing, watermarks and exactly-once sinks.
Thinking out loud.
Research, long-form and a Hinglish SQL series — the same ideas, pitched at whoever needs them.
A Survey of Trustworthy & Explainable AI
Accepted for presentation, 4–5 September 2026.
Your AI Can’t Explain Itself. Your Pipeline Is Why. ↗
Edition #1 — why explanation failures are usually data-layer failures.
Do Your Feature Importances Agree? ↗
Five attribution methods on one model — and an honest negative result.
Database kya hai aur Excel kyun kaafi nahi? ↗
SQL Zero to Interview #01 — a Hinglish series for working data engineers.
Where I've shipped.
Data Engineer · IBM
Optimized AWS Glue ETL to cut compute cost by 25%. Refactored PySpark jobs 25% faster, saving 200+ compute-hours a year. Built Kafka + PySpark streaming monitoring holding 99.9% uptime, and an enterprise RAG pipeline (LangChain + vector DBs) that cut manual review of financial documents by 40%.
Software Engineer, Data Engineering · Pine Labs
Migrated RDBMS workloads to Azure Data Lake and Databricks — PySpark pipelines over 5M+ daily records running 10x faster. Built a 100GB+ airline data warehouse on Azure Synapse consolidating 10+ vendor feeds, real-time Spark Streaming + Kafka pipelines, and 15+ production pipelines end-to-end.
Data Science Intern · NuvoRetail
Billing optimization and sales forecasting models hitting 90% accuracy — my first taste of data products that people actually bet decisions on.
PhD Scholar, Trustworthy & Explainable AI · VIT
Researching how to make LLM-era systems transparent, bias-aware and ethically compliant inside enterprise data platforms. First survey paper accepted at IEEE WorldSUAS 2026.
Where I talk data.
Explainable Pipelines
Weekly deep-dives where data engineering meets explainable AI — architecture patterns, trust checks, and the code behind them.
Bits to Insights
"SQL ratte nahi — samajhte hain." A Hinglish SQL series: concept → practical → interview questions, built for the next generation of data engineers.
pipeline.diaries
The diary of a data pipeline — "AI gets the credit. The pipeline keeps the receipts." One brutal truth from desi tech life, daily.
Long-form Notes
Newsletter editions and engineering write-ups, republished for readers who live outside LinkedIn.
Notebooks
Reproducible XAI experiments — one question, one experiment, one honest finding each. Negative results included.
Building in Public
Threads on what I'm shipping — open-source XAI tools, pipeline war stories, and research notes as they happen.
Explainable Pipelines
One idea a week on building data platforms that can explain themselves — pipelines, lineage, attribution, and the LLM era's trust problem. No spam, just engineering.
Subscribe on LinkedInLet's build something
worth trusting.
Open to collaborations in data engineering, explainable AI research, and technical content. The particles just spelled my name — now you know who to email.