Internal API gateway and control plane, managed through the OctoDash dashboard.

• Built and own Octopus, a config-driven internal gateway fronting ~400 internal and vendor integrations; adding or changing an integration went from a code deploy to a config change
• Sustained 99.99% availability over 30 months with zero Sev-1 incidents on a revenue-critical path: every vendor call in FinBox's lending journeys (e.g. PAN verification) runs through Octopus, which also meters client usage for billing via OpenMeter
• Built cost-aware vendor failover: a scorer ranks vendors on error rate, latency and cost from Prometheus, pulls failing vendors from the pool and restores them once healthy, cutting vendor-outage impact by 70% and saving ₹20L/month
• Reclaimed 10 TB of RDS storage and cut database costs by $24K/year by offloading bloated webhook and workflow JSON payloads from PostgreSQL to S3

In-memory Go workflow engine and JSON transformer.

• Built DataDancer, an in-memory Go engine for the CNCF-backed Serverless Workflow spec, for short-lived workflows that don't need Temporal's durability; runs 10M+ workflows/day across 4+ microservices with no orchestration infra to provision
• Made workflows fully config-driven: pluggable activities with built-in OpenTelemetry tracing, and a JSON transformation layer that remaps data across schemas and calls APIs mid-transform. This replaced hand-written mapping code and deploys, cutting new integrations from 3 days to 3–5 hours
• Migrated DataDancer's execution logs from PostgreSQL to ClickHouse, designing the schema around its query patterns (user/workflow/time sort key, bloom filter on run ID, ZSTD, 31-day TTL), freeing ~5 TB of RDS storage

Temporal-based no-code process engine, built with a team of five.

• Co-designed and built a no-code workflow platform: a Temporal-based engine for the same Serverless Workflow spec, plus a visual builder where 50+ application engineers publish business workflows without backend changes; runs 1M+ workflows/day and cut new journey launches from 3 weeks to 1 week
• Shipped every engine change without breaking in-flight workflows by preserving Temporal's deterministic replay, enforced through a replay-test suite run against existing workflow histories
• Built runtime guardrails (trigger-loop cooldowns, byte-level payload limits, cycle detection) so a misconfigured workflow throttles or terminates itself instead of starving other workflows

Multi-tenant configuration service.

• Extracted config management from the monolith into a standalone Go + gRPC service, replacing the duplicated, per-service config handling across 4 services with one shared platform
• Added config versioning and scheduled activation, so teams can stage changes in advance and have them go live at a set time without a deploy
• Built tenant isolation with namespace-scoped RBAC (read, write and namespace-admin permissions per user) and an audit trail of every config change

Multi-tenant webhook delivery platform with configurable retry strategies and dead letter queues.

• Multi-tenant webhook delivery with per-tenant configurable retry strategies and dead letter queue handling
• Pluggable queue backend — backed by SQS and EventBridge Scheduler, designed for easy replacement