Akshay Gupta

Merkle Science builds blockchain intelligence and compliance infrastructure used by financial institutions, exchanges, and regulators worldwide. I joined in 2021 as the youngest engineer on the team. By 2023, I was leading it.

The platform story: When I joined, the platform supported 7 blockchains and was under constant strain — unstable, expensive, and hard to scale. Over four years, I led five end-to-end architecture overhauls — each planned, executed, and delivered without breaking customer trust. Today, the platform supports 22+ blockchains, handles 500TB of data, runs at 99.99% uptime, and costs 80% less ($360K saved annually) despite carrying 3–4x the data volume. That’s not a one-time win. It’s a compounding pattern of doing more with less, repeatedly.

What I built:

• Harvester — a unified blockchain data extraction framework that reduced new chain integration time by 70%, turning a slow, bespoke process into a standardised, repeatable one. We went from 7 to 22+ chains.
• Nimbus — a real-time streaming pipeline (evolved from Apache Beam → Kafka + KSQL) that makes blockchain transactions available on the platform within 7 seconds of confirmation, at horizontal scale.
• TigerGraph overhaul — redesigned the entire schema and multi-hop network exposure algorithm from scratch. The result: P99 query latency under 1 second, at 90% lower infrastructure cost, with significantly better explainability for compliance teams.
• 4 ClickHouse migrations — each one standardising schemas, introducing pre-aggregated data models, and delivering consistent sub-second query performance at enterprise scale.
• Cost optimisation — led cross-functional initiatives with DevOps to keep GCP compute costs flat for 3 years despite 3–4x data growth, and orchestrated the move to ClickHouse Cloud to cut self-hosting overhead.

How I lead:

• Team & culture: Directly manage a team of 3 engineers while informally mentoring across Data Science, Product, and DevOps. Built a documentation-first culture where architecture decisions, runbooks, and hard-won learnings are team property — not locked in any one person’s head. Mentored junior engineers through structured code reviews and design guidance, helping them grow into independent contributors who own their projects end to end.
• Stakeholder management: Serve as the primary interface between data engineering and the rest of the business — translating complex technical constraints into clear business language and vice versa. Regularly present infrastructure decisions, cost trade-offs, and data quality insights to non-technical stakeholders, enabling informed decisions on resource allocation and vendor contracts. Led a company-wide initiative to redefine how transaction counts are measured — a deceptively complex project requiring alignment across Product, Data Science, and leadership before a single line of code was written.
• Navigating ambiguity: Consistently take ownership of high-stakes, open-ended problems where the path isn’t clear — whether that’s evaluating a new blockchain for feasibility, responding to an unplanned production incident, or deciding when to rebuild vs. iterate on aging infrastructure.
• Shielding the team: Proactively absorb on-call escalations and cross-team coordination so engineers can stay in deep focus. Pair this with deliberate knowledge distribution to ensure no critical system depends on a single person — including me.

Testbook is one of India’s largest ed-tech platforms for competitive exam preparation. I was one of the first two data hires — joining as an analyst, and leaving as a senior data engineer who had built the entire data function, a revenue-generating CRM platform, and ML-powered product features.

Starting from zero: I set up Testbook’s first analytics infrastructure — Excel to Jupyter notebooks to live Redash dashboards pulling from MongoDB, BigQuery, and MySQL. These became the operating layer for product, content, and marketing teams. One early win: identifying a post-signup dropout rate of 50% and reducing it to 30% through data analysis and targeted interventions.

The CRM that became a revenue engine: What started as a two-person checkout recovery experiment evolved into a fully custom internal telesales platform. I designed the lead flow architecture, database model, real-time ingestion pipelines (Pub/Sub + Cloud Functions), lead scoring, and attribution strategy. At its peak, this platform contributed to 30% of Testbook’s revenue — built from scratch because nothing off-the-shelf fit the use case.

The T-Score moment: Indian Railways exams use a statistical normalisation method called T-Score to rank millions of aspirants. When a major exam was announced with weeks — not months — to prepare, we saw an opportunity. I built the automation pipeline to compute and deliver personalised T-Scores to every user who attempted our test series, using population-level aggregates decoupled from user-facing delivery. We were the only platform in India offering this at scale. The result: 60% of aspirants for that exam came to Testbook.

Other highlights:

• Built a question difficulty quantification algorithm using Apache Spark — reducing content curation time from months to days and improving educational quality at scale.
• Architected a CDC-based MongoDB consolidation pipeline using Airflow and GCS, saving 50+ dashboards and 1000+ queries from breaking during a critical database migration.
• Founded and scaled the data engineering department to a 4-person team.