Outcomes, not adjectives.

Cutting a BigQuery bill by 57%

Problem. A Fortune 500 group's analytics spend on BigQuery was growing faster than the value it produced — full-table scans, inefficient joins, unpartitioned queries, and no cost governance.

Approach. I led a FinOps engagement that profiled the heaviest queries, refactored anti-patterns (partition/cluster pruning, materialized views, MERGE optimization), and moved the workload onto slot-based capacity with a cost-governance model the team could run. I also built an analyzer that flags SQL anti-patterns automatically, so the discipline outlived the engagement.

“Together, we undertook a strategic cloud cost optimization and financial governance program that resulted in a significant reduction in overall BigQuery warehouse costs. What stands out about Pratik is his ability to work as a true partner.” — Ashish Chandra, Head of Data Engineering at Tata Steel (recommendation)

A 30K+ TPS transaction engine on Kubernetes

Problem. Telecom/FinTech partner integrations needed a transaction platform that could sustain tens of thousands of transactions per second without double-charging or losing money on failures.

Approach. Leading a team of ten engineers, I designed an idempotent, horizontally-scaled processing pipeline on Kubernetes with ledger/reconciliation, dual-layer auth (JWT + API keys), PCI-aligned data protection, and error-code orchestration so partial failures resolved deterministically.

Serving 1M+ users and cutting latency 47%

Problem. A social network backend serving over a million users had grown into a sprawl of services with slow, inconsistent API response times.

Approach. I designed Go/gRPC microservices behind a GraphQL gateway, consolidated services, and moved inter-service contracts to protobuf — cutting API response latency by nearly half while leading a remote-first team of four.

Production agentic AI for the UK energy grid operator

Problem. The National Energy System Operator needed agentic AI moved from promising prototype to something dependable enough for critical national infrastructure — where governance, auditability, and trust are non-negotiable.

Approach. Embedded as a forward-deployed engineer, I designed and hardened multi-agent AI into production with a strong emphasis on AI governance and compliance — the architecture documented as a set of interactive diagrams in my architecture portfolio.

A compliant Open Banking platform across ADGM, DIFC & SAMA

Problem. An Open Banking platform for the UAE and Saudi markets had to satisfy three regulatory regimes (ADGM, DIFC, SAMA), with data residency and consent management — while still being fast.

Approach. I built the consent-management and data-residency architecture and a retrieval-augmented semantic search layer on AlloyDB AI, cutting query latency 37% while keeping data within jurisdiction.

Minimal-downtime migrations to Cloud Spanner

Problem. Google's open-source Spanner migration tool needed stronger backend APIs and change-data-capture pipelines so customers could migrate to Cloud Spanner with minimal downtime and good post-migration performance.

Approach. As a core contributor during a Searce engagement, I shipped backend APIs for the schema assistant and built CDC pipelines on Datastream → Pub/Sub → Dataflow, plus primary-key and interleaving guidance that cut post-migration query cost. The track continues: 12 further upstream fixes in review from my personal account as of late 2026.

“I had the pleasure of working with Pratik on a Spanner migration tool, an official Google Cloud product, where he was central to its success. Pratik led a team of four engineers […] and contributed meaningfully to the pre-sales effort.” — Anupam Bhattacharyya, who managed the engagement (recommendation)