Lead Data Engineer · IIT Guwahati · Bengaluru, India

I build data platforms that handle scale most teams never see — and keep them running while rebuilding them from the inside out.

At Merkle Science, I led 5 full architectural overhauls of a blockchain compliance platform — taking it from 7 chains to 22+, 500TB of data, 80% lower costs, and 99.99% uptime, with zero customer disruption. None of those overhauls came from chasing new technology. They came from understanding the system deeply enough to see what was actually broken, and fixing exactly that — usually with something simpler than what was already there.

Before that, at Testbook, I was one of the first two data hires. I built the analytics stack from zero, designed a telesales CRM that drove 30% of company revenue, and shipped the only platform in India delivering personalised T-Scores at scale — capturing 60% of aspirants for a major Indian Railways exam cycle.

Harvester — Unified blockchain data extraction framework. Reduced new chain onboarding time by 70%. Scaled from 7 → 22+ chains.

Nimbus — Real-time streaming pipeline (Beam + Pub/Sub → Kafka + KSQL). Blockchain transactions on-platform within 7 seconds of confirmation.

TigerGraph overhaul — Redesigned graph schema and multi-hop network exposure algorithm from scratch. P99 latency < 1s at 90% lower infra cost.

ClickHouse migrations (×4) — Standardised schemas, pre-aggregated models, consistent sub-second query performance at enterprise scale.

Telesales CRM @ Testbook — Built end-to-end: lead flow, DB modelling, Pub/Sub + Cloud Functions pipelines, lead scoring, attribution. Peaked at 30% of total company revenue.

Languages      Python · SQL · Scala · Go · Java
Databases       ClickHouse · TigerGraph · BigQuery · PostgreSQL · MongoDB · Redis
Streaming       Apache Kafka · KSQL · Google Pub/Sub · Apache Beam
Orchestration   Apache Airflow · Apache Spark · Kubernetes · ArgoCD
Cloud           GCP · AWS · ClickHouse Cloud

• Simplicity is earned, not chosen — every overhaul I've led came from understanding the system deeply enough to see what actually needed fixing. The simple solution at the end is the reward for that understanding, not a shortcut past it.
• Reach for boring first — new technology is a cost. I'll pay it when the problem genuinely demands it, not because it's interesting. Most problems don't need a new tool; they need a clearer model of what's already there.
• Owning hard problems end to end — I don't hand off the ambiguous parts
• Documentation-first culture — architecture decisions and hard-won learnings should be team property, not locked in someone's head
• Mentoring that actually sticks — structured code reviews, design guidance, helping engineers own their work fully

🌐 akshaygpt.in · 💼 linkedin.com/in/akshay-iitg · 📬 akshay.iitghy@gmail.com