Shashank Adari

In plain words

A small tool that proves a copied table is identical to the original. When a company moves data between systems, "the row counts match" isn't enough: two tables can have the same number of rows and still disagree. This tool fingerprints every row on both sides and produces a report an auditor can file.

What it does

• Catches a lost row and a duplicated row even when the counts still match
• Ignores harmless differences: NULL vs empty, CHAR padding, letter case, INT vs DECIMAL vs DOUBLE
• An order-independent fingerprint per table, so two teams can compare one value without sharing data
• Markdown and JSON reports; a Databricks notebook that files the evidence and stops the pipeline on failure

The technical details

Compares a legacy table with its lakehouse copy on three checks: schema comparison, row counts, and a SHA-256 row-hash reconciliation compared as a multiset, with variance thresholds. A Spark engine runs on Databricks; a pure-Python engine applies the same rules to small extracts and runs the test suite, and a parity suite proves both produce byte-identical hashes.

## fact_sales: NOT SIGNED OFF

| Check | Result | Detail |
|---|---|---|
| Schema comparison | PASS | extra in target (allowed): load_ts |
| Row counts | PASS | source 5, target 5 (0.0000% variance) |
| SHA-256 row hashes | FAIL | 2 source rows without an identical target row (40.0000% variance) |

Built with

In plain words

A website that teaches the fundamentals of programming (data structures and algorithms) through 31 short, interactive 3D stories set in everyday Indian life: a lost page at a wedding hall, a tiffin ladder, a cricket scorer, the last bus home. Each story shows a real problem, a first attempt that fails, the right approach step by step, the traps, and practice questions.

What it does

• 31 interactive 3D episodes, the first two free
• Sign-in with one active device per student; progress saved to their account
• Members-only pages served encrypted with a fresh key each time, plus a visible watermark
• A readiness check, a mistakes notebook, badges and printable certificates

Some of the episodes

The technical details

Alongside the episodes: two study guides (DSA and Data), 20 SQL exercises on a deliberately messy dataset, and a "Pattern Gym" of 45 interview-style problems. Students sign in, their progress follows them across devices, members-only pages are encrypted per visit and watermarked, and each season ends with a certificate. A narrator reads every chapter aloud using the device's own voice. Hosted on Cloudflare Pages with Supabase, built by a Python script that checks everything agrees before it publishes.

Built with

In plain words

This site is built the way I build data platforms. Every night it pulls my GitHub activity, keeps a history that never ages out, checks the data, and republishes itself. A new blog post is just a text file; the site picks it up and goes live within minutes. I designed and built it together with Claude Code.

What it does

• Daily GitHub tracking with streaks and history
• Blog from Markdown files, with scheduling and drafts
• A read-only SQL console over the site's own data, for curious visitors

The technical details

A Python pipeline with bronze, silver and gold layers, data-quality expectations that fail the build when content is wrong, and a GitHub Actions deploy to GitHub Pages. The architecture diagrams on this site are generated as SVG at build time from a small layout file.

Built with

The technical details

1. Designed and built a config-driven Unity Catalog → ADLS Delta sync framework (Databricks notebook, JSON table registry, ADF orchestrator) with full-refresh and MERGE modes, parallel fan-out, run-overlap protection, dry-run mode and tiered verification.

   51B rows · 1h window

2. Built the migration sign-off framework: schema comparison, row counts and SHA-256 multiset row-hash reconciliation across Databricks workspaces with variance-threshold gating, producing audit-ready evidence for production approval.

   audit-ready

3. Converted complex nested Teradata views and staging tables to Spark SQL and PySpark, re-pointing acquisition-layer views to enterprise foundation master data through hash-based join keys, with production grant standards and automated QA checks.

4. Defined the Unity Catalog and Synapse migration playbook: governed external-location access, clone-versus-CTAS copy strategy, Delta transaction-log recovery from checkpoints, and T-SQL → Spark SQL semantic-parity rules.

5. Leading the split of one Azure Data Factory into two regional factories (UK and US); re-engineered ARM template parameterization to unblock CI/CD and drive PreProd/Prod configuration parity.

   739 → ≤256 params

6. Built decommissioning and FinOps tooling: a schema-usage audit over Databricks system tables plus workspace, job, pipeline and dashboard scans; a cross-repository ADLS dependency inventory; and Azure run-cost models from rate-card pricing and pipeline telemetry.

7. Led AI adoption for project teams: evaluated frontier LLMs, then rolled out LLM-assisted code generation, review, audits and documentation with guardrails (restricted Databricks data access, verification rules, dry-run defaults, human review).

Tools and techniques

The technical details

1. Analyzed BTEQ and KSH jobs and Teradata DDL, implemented Delta models in a layered medallion architecture, and built PySpark and Spark SQL pipelines with hash-based surrogate keys and error-quarantine tables.

   4 regions

2. Built PySpark MERGE pipelines for SAP BW-origin tables and JDA / Blue Yonder supply-chain feeds, with automated staging validation and per-step freshness and volume audits.

3. Replaced KSH and BTEQ orchestrators and trigger-file generation with Databricks Workflows backed by Delta control tables, plus ADF and Event Grid gatekeeper pipelines that confirm every load in a dependency set through the ADF REST API before publishing completion events.

4. Hardened ADF CI/CD release automation with PowerShell retry logic (exponential backoff, pre-flight gates) for control-plane eventual consistency, and root-caused cross-environment drift in trigger and Event Grid configuration.

5. Resolved count and type discrepancies from Teradata-to-Spark semantic differences (case sensitivity, numeric overflow) and Delta write concurrency, with documented root-cause analysis.

Tools and techniques

The technical details

1. Built real-time telemetry pipelines on Azure IoT Hub, Event Hubs and Stream Analytics into Azure Data Explorer (Kusto) and SQL databases, powering a real-time manufacturing monitoring platform.

   −35% material wastage

2. Developed batch ETL in Azure Data Factory and automated alerting in Azure Logic Apps; built Angular dashboards and Power Apps front-ends for production users.

3. Led an Azure PaaS optimization initiative; automated PaaS deployments with Python and monitored the estate with Grafana and Datadog.

   −20% cost · +30% perf

Tools and techniques

The technical details

1. Resolved 700 incidents and service requests within SLA, documenting root cause in ServiceNow to prevent recurrence.

   700 within SLA

2. Bridged business stakeholders and the development team, turning requirements into delivered changes; contributed to code reviews and knowledge-sharing sessions.

Tools and techniques