Solutions Architect, Chief Data Office · Data Governance · Data Engineering · AI Governance
Fifteen years turning messy enterprise data into something people — and now AI agents — can actually trust and act on. I work at the join between governance, engineering, analytics and architecture, because that is exactly where the value hides.
The consolidated view
Four disciplines around a single centre. Governance, engineering, analytics and AI are usually run as separate programmes — the value shows up exactly where they meet. Node size shows depth; the outer ring is the leadership layer that holds the four together. Hover any node, or filter by discipline.
What I do
Each one is a job on its own. Run together, they are the difference between a number you can defend and a number you argue about.
Pipelines that are safe to re-run at 3am. Incremental ingestion with watermarks, SCD Type 2, declarative quality gates, and lakehouse foundations across Snowflake, Databricks and Microsoft Fabric — federated with Starburst/Trino where the data cannot move.
Federated ownership, stewardship and data products with named owners and SLAs. Councils that decide, working groups that execute. Glossary, lineage and quality wired through Collibra with the teeth to stop a load — not living in a slide deck.
Certified data sources, standardised KPIs, row-level security and tuned extracts — governed self-service instead of conflicting numbers. Plus the unglamorous half: administering a 500+ workbook estate, rationalising it, and migrating legacy Cognos and Excel onto something maintainable.
Agents built in Agent Studio with Prompt Builder — topics, actions and guardrails — grounded through Data 360 over Snowflake, so leadership can ask a question in plain language and get an auditable answer. An agent is exactly as trustworthy as the certified datasets, entitlements and lineage beneath it.
“Every organisation wants AI agents acting on its data. Very few have data an agent should be trusted with. The unglamorous work — ownership, stewardship, quality rules, lineage — is suddenly the thing that decides whether AI helps or quietly multiplies your existing errors.”
— and the reason all four disciplines above are one job, not four.How it fits together
Three diagrams for the three questions I get asked most: what does the platform look like, who is actually accountable, and how did someone end up doing all four of these things.
The shape I build towards. Note where governance sits: not a stage in the pipeline, a rail underneath all of it. That is the whole argument.
Governance thrives when roles are genuinely defined. Owners, stewards, custodians and consumers each hold a distinct piece; when one is vacant, the chain breaks there and only there.
Each layer starts when I moved into it and then just keeps going. The direction of travel is downward — analysis and visualisation first, then analytics, then the engineering, architecture and governance underneath them — but nothing above ever gets handed back.
Experience
From Tableau developer to Chief Data Office. The titles changed; the job — making data trustworthy enough to decide on — did not.
Pune, India · Hybrid
Defining target-state BI and data architecture, turning business and regulatory requirements into governed, decision-ready reporting. Solutions Architect working deep in the Salesforce ecosystem.
Self-employed
Maharashtra, India · Remote
Governed Tableau architecture, migration and reporting for asset management, banking, healthcare and product-engineering clients — each engagement run end to end.
Bengaluru, India · Hybrid
Tableau Design Architect delivering credit-risk analytics for Consumer & Corporate Banking in a highly regulated environment.
Pune, India · Remote
Led delivery for retail and software reporting across multiple concurrent workstreams.
Pune, India · On-site
Owned the full BI lifecycle for global supply chain and procurement in Emerson's Core Data Analytics CoE.
Pune, India · On-site
Predictive analytics for a major U.S. insurer.
Pune, India · On-site
Tableau trainer and developer building dashboard capability for customer-experience programmes.
DenuoSource India
Hyderabad, India · On-site
Retail and pharma analytics for Walmart and Office Depot.
GateForum
Jabalpur, India · Remote
Market and competitor research in the equity crowdfunding sector — SWOT analysis, survey design and insight-driven recommendations that shaped early-stage investment approaches.
Certifications
Each of these maps to something on the job — not collected for the wall. Badge artwork is the real credential where I hold it. Cards link to the issuing programme; the Microsoft credential links to my verified transcript.
Skills & stack
The full list behind the capability map. Highlighted items are the ones I am hands-on in this quarter, not ones I touched once in 2017.
Hands-on since 2016 · production-grade since 2018
Multi-cloud — Azure primary, Snowflake as the constant
The centre of gravity — CDMP Practitioner, Collibra in production
Since 2012 · 5× Tableau certified · 500+ workbook estates
Predictive since 2016 · agentic since 2024 · IAPP AIGP
The layer that wraps the other four
Data platforms are only as good as the domain understanding behind them.
CDO reporting, AUM and net-flow metrics, mandate and benchmark data, investment-risk and controls reporting, regulatory (FED) submissions.
Credit-card and corporate credit-risk estates — exposure, delinquency, vintage, loss forecasting and portfolio-risk reporting.
Purchase-to-pay and SAP procurement data, inventory, logistics, vendor performance and on-time-delivery analytics.
Market-basket analysis, cross-sell and loyalty segmentation, competitor price indices for dynamic pricing, store-traffic prediction.
Policy and claims analytics, predictive risk modelling from claim, location and climate data.
Revenue, cost and margin dashboards; operational and smart-meter style data; network and customer-experience reporting.
Open source
Eight reference implementations of the patterns on this page. All of it was executed before it was published —
the pipelines run, the tests pass, and the numbers in each README came from a real run rather than an estimate.
Clone any of them and run pytest.
A banking domain data architecture worked end to end. Two functions read the same book of derivatives and report four different numbers — none of them wrong. One canonical model, a finance lens and a risk lens, a published reconciliation, and 42 data standards, ten of them enforced by a linter that fails the build.
A working Data Vault 2.0 over real SAP procurement tables — LFA1, EKKO, EKPO, EKBE, MARA. Four source systems, hubs / links / multi-source satellites, hash keys and diffs, PIT + bridge, business vault and star marts. Reproduces the bug where SAP purchase-order numbers collide across instances because they are not globally unique.
Medallion architecture in PySpark that runs on a laptop with no cluster: incremental ingestion with watermarks, SCD Type 2, a declarative YAML data-quality engine and as-of dimensional joins. Includes the future-dated row that poisons a watermark so a job silently ingests nothing, forever.
What actually happens on the cluster: map-side combine, shuffle bytes, partitioning, caching and skew — each one measured, then shown as the DataFrame/SQL equivalent you should actually ship. A lab, not a lecture.
Config-driven connectivity across MySQL, Postgres, Oracle, Snowflake, SQLite and ADLS behind one interface. Secrets never in the repo, connections that always close, bulk loads that batch, retries that classify errors properly.
The working artefacts of a governance function — DAMA-DMBOK aligned glossary templates, a 61-rule quality catalogue with a runnable YAML-driven engine, stewardship operating model, RACIs, a CDO KPI framework and a lineage guide. Includes the suite that reported 100% pass on broken data because NULL > 0 is NULL, not false.
Grounding enterprise AI agents in governed data — retrieval quality is a data-governance problem, not a prompting problem. Reference architecture, the semantic layer as a metric contract, guardrail patterns, AI controls mapping and a runnable evaluation harness.
Enterprise Tableau standards that keep a large estate healthy: certified data sources, row-level security patterns, performance tuning, content rationalisation, Server→Cloud migration runbooks, plus read-only estate-audit, Hyper API and TabPy scripts.
Three of these reproduce failure modes I have actually hit and fixed. Reproducing a bug is worth more than describing one — and it is the fastest way for you to check whether I know what I am talking about.
Knowledge base
I keep a running, structured record of every session of the cloud and data-engineering programme I'm working through — appended into living documents rather than scattered files, so it stays readable end to end. Below is the map of it. Where a topic is worth proving, it becomes a repository.
Writing & speaking
Selected pieces on governance, BI and the agentic era — with the reach they earned. Published on LinkedIn, where I also post most weeks.
Presented at a governance forum: people, data, data stores, processes and governance. The point that landed hardest — governance thrives when roles are genuinely defined. Owners, stewards, custodians and users each hold a distinct piece of the trust chain.
The next breakthrough isn't another framework — it's the shift from documented governance to active, metadata-driven governance. Data discovered automatically, sensitive fields classified in real time, lineage tracked continuously, audit evidence generated as part of the workflow rather than after the fact.
AI can detect anomalies, suggest classifications, summarise lineage and draft audit evidence. What it cannot do is own accountability. The future isn't AI-driven governance — it's human-led governance, supported by AI-enabled intelligence.
The question is no longer “can we visualise our data?” but “can an AI agent trust our data enough to act on it?” Azure for infrastructure gravity, a semantic trust layer in the middle, agentic analytics on top. Teams still debating which BI tool are answering last decade's question.
“Agents are only as trustworthy as the data beneath them. Ingest high-quality, well-governed data and the agentic layer becomes genuinely reliable. Skip that foundation, and you've simply automated your data problems.”
— from the modern data architecture seriesGet in touch
Open to conversations on data governance, data engineering, BI leadership, AI governance and enterprise data architecture — roles, advisory work, collaboration or speaking.
Based in Pune, working with teams across India, Europe and the US. On-site, hybrid or remote.