I lead engineering for an AI-native, multifamily property-management platform, taking it from early architecture to a production system used by real operators. I designed the platform’s multi-tenant, event-driven core with fine-grained authorization, directed the AI agent stack through three architectural generations, and built a tenant-aware knowledge/retrieval system and a multi-provider voice AI layer on top of it. I also built and run the team delivering it, scaling to 15 engineers organized into module pods.

Within Paytm’s Promotions team, built and scaled order-processing systems spanning promo codes, single- and multi-item bank offers, and EMI offers — a combined 10 million orders a day. Also drove performance and reliability work across the team’s four-datastore stack (MySQL, Elasticsearch, Cassandra, Redis): redesigned a bulk promo-code generation pipeline to remove its materialized-view dependency and unblock a Cassandra 4 upgrade, pushed rollup aggregation into Cassandra to cut a transaction pipeline’s network and processing overhead, cut Elasticsearch query latency 50%, and introduced TestContainers as the team’s DAO-layer testing standard.

Joined as a founding engineer on the Financial Reports team and built the v0 reporting pipeline (Airflow, Kafka, Postgres materialized views), then led it through two scaling redesigns — splitting read and write into separate services, then replacing full materialized-view refreshes with asynchronously computed daily/quarterly/rolling Postgres aggregate tables that cut API latency in half. Introduced Architecture Decision Records and brought integration-test execution into the team’s Kubernetes-based deployment pipeline.

Held technical leadership for two products — IMConnecting and IMSecurities — spanning architecture, technology evaluation, prototyping, RFPs and pre-sales. Architected a self-serve, Elasticsearch-backed dashboard platform letting customers compose their own queries and dashboards. On IMConnecting, designed for two opposite load shapes within the same election-processing system — a high-concurrency retail path (Redis and beginning-of-day caching plus async submission, cutting API latency 90% with 4x throughput) and a high-fan-out institutional bulk-submission path handling up to 35k accounts per request via chunked, parallelized processing on ActiveMQ.

Took IMConnecting, a brand-new financial product, from 0 to 1 — a Spring/Postgres/ActiveMQ web app with two variants: an issuer side for stock exchanges and an investor side for election submissions. Built the issuer variant’s ~70 form types on a shared Model-Driven Development template (a reusable template plus lifecycle rules, an in-house code generator), giving consistent design and validation, a 25% lift in team productivity, and standardized code by construction. Owned the product’s non-functional foundations across both variants — support for multiple authentication schemes (RDBMS, LDAP, SSO, PKI), API-level authorization, and a system design weighted toward reliability and ACID guarantees given the product’s financial nature.