Vignesh Kumar A

Zoho Corporation · Chennai

• On the distributed database team, built a read-only consistency verifier for the cluster, modelled on PostgreSQL's amcheck (50 checks, 17K lines of C and SQL). It checks shard placement, colocation, replication residue and two-phase commit state on every node over asynchronous libpq with a deadline per node, fingerprints each finding so a fault keeps its identity across scheduled runs, and proves every check with a fault-injection test on a 10-node regression cluster.
• Rewrote the operation layer for range-partitioned shards in C, folding 2,800+ lines of user-defined function code into one operation model with cutoff expressions and operation-aware scheduling, and gave every range operation a dry run that shows what it would change in live metadata before a single shard is mutated.
• Extended the team's blue-green query validator (Java), used to test major-release upgrades, to export each coordinator's planning and execution time for every validated query, tagged by release for query-by-query comparison.
• Built the verifier's test and tooling stack: a tiered test framework (unit, 10-node integration, upgrade and coverage) behind one CI gate, a build-time lint that rejects any SQL with side effects, and a Python MCP server so engineers can run checks and triage findings from an AI assistant.
• On the point-of-sale platform, designed and built the event-driven sync layer in Java on Apache Kafka and MySQL: one event per change instead of M×N API calls, which cut upstream API load by 99% and brought new-service onboarding from weeks to under 5 minutes through a declarative framework adopted by multiple teams.
• Made sales durable across server crashes with a transactional outbox written in the same MySQL transaction as the sale, with background retry: 99.9% consistency at 1K+ transactions per minute between the registers and the finance platform.
• Split the distributed cache (JVM + Redis) into isolated segments with version-based invalidation, raising hit rate from 84.3% to 98.1% and removing 131M database queries per week.
• Delivered real-time push to 1,000+ concurrent WebSocket endpoints, cutting data conflicts by 98%; designed a three-level circuit breaker that opens within 15 seconds of downstream slowness; and built correlation-ID tracing across asynchronous threads that cut root-cause analysis from 3+ days to minutes.
• Wrote the team's code-review guidelines, published 12+ internal engineering blog posts, and mentored 3 engineers.