Zeel Kakadia

Authored self-contained statistical evals end to end — framing the dataset and analysis contract, implementing the oracle solution that defines ground truth, and building the scoring verifier plus hidden test variants.

• — Authored statistical evals end-to-end, each owned from dataset to verifier — mostly R, with Python/Colab and Julia.
• — Survival analysis (Cox PH, Kaplan–Meier), mixed-effects / multilevel models, multiple-testing / FDR, Bayesian & empirical-Bayes shrinkage.
• — Econometrics & causal inference (Heckman selection, propensity matching, ordered logit); unsupervised methods (LDA, MDS, EM mixtures, factor analysis); spatial statistics; ML calibration / drift / conformal prediction.

Built the hardest tier of evaluation tasks on real, canonical datasets — genuine applied statistical analysis, not toy data.

• — Built the hardest-tier tasks in R (with some Julia) — survival, panel, econometrics, robust and mixed-effects methods.
• — Examples: mortgage-lending data → logistic discrimination; labor-supply data → Heckman selection; investment panels → dynamic panel GMM (Nickell bias).
• — Clinical survival data → Cox PH diagnostics + competing risks; overdispersed counts → GAMLSS dispersion modeling; repeated-measures data → mixed-effects / AR(1) / NLME family.

Authored register-transfer-level (RTL) hardware evaluation tasks for frontier-model benchmarking.

• — Built RTL hardware tasks used to benchmark frontier models on hardware-design work.

Onboarding every new customer meant moving their entire MySQL + Neo4j footprint onto a new architecture — slow, manual and heavily reliant on DBAs. I automated the whole thing end to end.

• — Automated customer-project migrations across MySQL & Neo4j with minimal DBA involvement.
• — Backed up MySQL (MySQL Shell) and exported Neo4j graphs (Cypher-shell), including source-environment cleanup.
• — Built a GitHub Actions CI/CD pipeline to back up, upload to version-controlled repos, and restore.
• — Versioned every schema change with Flyway for a traceable, consistent migration history across environments.

The platform’s graph data had to respect complex, per-role access rules. I built the service that generates the right queries dynamically and provisions each project’s database automatically.

• — Built a graph backend on Neo4j + MySQL with dynamic Cypher generation driven by role-based access filters.
• — Developed an automated provisioning system that creates schemas, users and role configs per project deployment.

Made the platform observable and resilient — so we always knew who changed what, and unhealthy services were caught before customers ever noticed.

• — Designed a real-time Audit Framework tracking user and role modifications.
• — Built liveness & readiness probes for MySQL, Kafka, Neo4j and Redis to monitor application health.
• — Replaced the deprecated PhantomJS with Puppeteer (headless Chrome) to automate chart-snapshot generation.

Connected the data platform to its AI features and to geospatial search — turning raw metadata and map data into something models and queries could actually use.

• — Generated batch-wise JSON of DataView metadata (measures & dimensions) for entire projects, auto-uploaded to MinIO to feed a GenAI metadata push.
• — Extracted OpenStreetMap City/State/Country boundaries (multipolygon geometry) and indexed them into OpenSearch for fast, accurate custom-border geospatial queries.

Raised the reliability bar — hardening the test suite and resolving the issues that actually mattered in production.

• — Wrote 80+ JUnit tests and API automation suites, systematically resolving critical failures.
• — Resolved production bugs and shipped performance optimizations, refactoring for code quality and maintainability.