Bengaluru, India·Available for high-impact systems roles
01 // Experience
Jun 2022 – Present
Senior Software Engineer
Hevo Data · Bengaluru
Jun 2022 – Present
:: Agent-Native Tooling
–Built the surface that lets LLMs generate and operate connectors: construct library, connector generator, installable CLI, skill library, and an MCP server exposing platform APIs to agents.
–Started the internal agent-tooling platform the team now builds on — repo structure, skill conventions, CI auto-release, and marketplace distribution.
:: Connector Platform & Architecture
–Made connector authentication declarative: transformed auth from bespoke platform processors, DB migrations, and cross-service coordination into zero-code configuration files.
–Moved runtime configuration and config templates into connectors behind a compatibility kit that gates correctness at build time rather than in review.
–Standardised connector repo creation, versioning, and release behind a central BOM, decoupling the fleet from the shared platform SDK runtime.
:: Connection Reliability & Validation
–Owned the test-connection contract end-to-end; dropped false-negative rate from >30% to nil in validation — ensuring failing checks indicate true connection faults.
–Delivered pre-flight configuration validation across all sources/destinations (Postgres, MySQL, Oracle, SQL Server, Snowflake, BigQuery, Redshift) with p95 latency < 10s.
:: Failure Visibility & Performance
–Engineered dynamic error classification, cutting classification cycle time from 3–7 days (full release) down to minutes.
–Root-caused production pipeline hangs to the regex engine; migrated to RE2J to bound worst-case evaluation under 5s, reducing load test stalls from several minutes to 1m19s.
–Instrumented connection health metrics and HTTP client telemetry in construct libraries to detect API rate-limiting before sync failure.
:: Data Correctness & Scale
–Delivered SCD Type 2 history mode across Snowflake, BigQuery, and Redshift (including Redshift's non-MERGE four-statement path).
–Implemented real-time CDC on Debezium and built the schema catalog service from scratch — schema versioning, metadata management, and compatibility checks across evolution.
–Scaled object handling from 1,000 to 25,000 objects with a 5x gain in MySQL ingestion throughput, and engineered parent-child handling with inferred deletes.
:: Engineering Velocity
–Cut build/CI runtimes against 90-day baselines: mvn clean verify from 31:43 to 10:33, CI from 43:32 to 21:24 on initial service, and 11:07 to 5:15 across the fleet.
–Built the unified developer CLI and local platform so teams execute full pipeline stacks locally rather than contending for shared cloud environments.
–Engineered a local, read-only MySQL/MariaDB binlog viewer, decoder, and transaction analyzer.
–Indexes decoded binlogs in SQLite with a virtualized event stream, transaction grouping, row-image inspection, and anomaly detection across 6 position traps.
–Built an interactive TUI/CLI for running AI coding assistants (Claude Code, Codex, Antigravity) in isolated profiles with separate configs, token states, and histories.