Mallika Chunduru MallikaChunduru

I make numbers that executives bet on, and I can show my work.

Four years turning pharmaceutical and healthcare data into dashboards, KPIs and answers, at AstraZeneca and Cigna.
I care most about the unglamorous part: whether the number is actually right, and whether anyone can prove it.

🧭 The arc

Not a career change. A change in altitude.

2021 ──────────── 2023 ──────────── 2025 ──────────── now
  │                 │                 │                │
  Cigna, India      MSc Data Science  AstraZeneca, UK   building in the open
  Healthcare        Teesside Univ.    Pharma commercial
  claims & ops      (Advanced         BI for 700+ users
  reporting          Practice)
  │                 │                 │                │
  "make the        "learn why the    "make the        "prove the number
   report"          method works"     number trusted"   in public"

I started where most analysts do: building the report someone asked for. Healthcare claims at Cigna taught me that the hard part was never the chart, it was the reconciliation underneath it, and that a discrepancy nobody chases becomes a decision nobody should have made.

An MSc in the UK gave me the statistical vocabulary for what I had been doing by instinct. At AstraZeneca that turned into ownership: KPI definitions, dimensional models and validation gates for reporting that 700+ people read without ever seeing my name on it.

The pivot happening now is from report author to analyst who engineers. Same domain, higher standard: version control, tests, measured results and documented trade-offs, which is what the repositories below are for.

🎯 Currently

At work

Power BI dashboards over Snowflake and Azure Synapse for commercial and operational KPIs
Reusable DAX measures and dimensional models so a KPI means one thing everywhere
Source-to-target reconciliation on ADF and Databricks deliveries before anything reaches a report
Root cause analysis on reporting discrepancies, with the business and data engineering teams
Copilot, Azure OpenAI and Power BI AI for report summarisation, kept on a short leash

In the open

Data quality as code: dbt tests for NHS numbers, ICD-10 and claims duplicates
Snowflake cost attribution, because invoices never say which query is expensive
DAX performance antipatterns, benchmarked instead of asserted
Tableau to Power BI migration tooling, from having done the migration by hand
Every result in those repos is measured by committed code, not claimed

🧱 Building blocks

📊 Where I actually sit on the stack

Levels are claims I can defend in an interview, so each one names where it was earned.

Capability	Depth	Earned on
SQL (window functions, optimisation, reconciliation)	`████████░░` Advanced	Daily at AstraZeneca and Cigna; query tuning for recurring reporting
Power BI & DAX (models, measures, report performance)	`████████░░` Advanced	700+ user dashboards; Tableau to Power BI migration at Cigna
Excel, Power Query (M), VBA	`████████░░` Advanced	Automated recurring report packs, cutting manual preparation
Data validation & ETL testing	`███████░░░` Strong	Source-to-target reconciliation on ADF and Databricks deliveries
Python (Pandas, NumPy, analysis)	`███████░░░` Strong	Trend, forecasting and validation work; the repos below
Snowflake	`███████░░░` Strong	Primary warehouse for commercial reporting at AstraZeneca
Dimensional modelling	`███████░░░` Strong	Standardising KPI calculations across report suites
Azure data platform (ADF, Synapse, Databricks, Fabric)	`██████░░░░` Working	Validation and consumption side, alongside data engineering
Statistics & forecasting	`██████░░░░` Working	MSc Data Science; variance and trend analysis in role
Tableau	`█████░░░░░` Working	The source system in the Cigna migration

📦 Selected work

Built in the open, with tests, CI and measured results. Every number below is produced by code committed in that repository.

dbt-clinical-dq

dbt SQL DuckDB

Healthcare data quality as dbt tests: NHS number checksums, ICD-10 format, duplicate claims, staging to mart reconciliation.

1.0 detection recall over 160 labeled violations, 0 false positives.

claims-denial-radar

Python SQL scikit-learn

Predicts claim denials before submission, with a SQL feature layer built so target leakage is structurally impossible.

38.5% lower simulated handling cost; threshold chosen by dollars, not accuracy.

dax-perf-patterns

DAX Power BI Python

Ten benchmarked DAX antipatterns plus a static analyser for measures and semantic models.

Scanner reports its own 1.00 precision and recall, gated in CI.

twb2pbi-migration-kit

Python Tableau DAX

Turns Tableau workbooks into a Power BI migration plan, from having done that migration manually.

88.5% auto-mapped calculations; a 2,246-calculation corpus in 0.32 s.

snowflake-spend-lens

Snowflake SQL FinOps

Finds where Snowflake credits actually burn, using literal-stripping query fingerprints.

1M queries attributed in 3.7 s; runs offline against a committed ACCOUNT_USAGE replica.

medallion-kpi-lakehouse

PySpark Fabric Databricks

Bronze, silver and gold with contracts at every boundary and re-runs proven to be no-ops.

Every row accounted for: 1.22M rows, conservation asserted per batch.

→ All 12 repositories

🧪 How I work

🔍 Trace it before you trust it. A KPI without a reconciliation path is an opinion with a number attached.

📐 Define it once. Reusable measures and models, so "revenue" cannot quietly mean three things.

📉 Report the boring result. If the simple baseline wins, that goes in the README too.

🗣️ Write for the reader. Stakeholders need the decision, not the query plan.

☕ Outside the code

📂 open healthcare data · 🎨 dashboard design craft · 📚 statistics that survive contact with real data · ✍️ writing up data quality war stories

🤝 Quick connect

Open to Data Analyst and BI roles in the UK, particularly in healthcare, pharma and other regulated data environments.

Navigation Menu