Harikrishnan G
Backend Engineer · 5 years
$ ls skills
$ ls experience
Designed a reusable upsell framework (pluggable trigger evaluators, lifecycle tracking, priority cascading) powering C&O Pro, Cardio & AFib upsells driven by RHR, HRV, sleep and cycle data; conversions attributed via placement-based UTM tracking.
Personalized BetterHelp content surfaced from logged emotional tags. Intent-tiered 3-article journeys, gender-scoped configs, state-machine interaction tracking, and per-day frequency capping.
YAML-driven retest upsell with weighted biomarker scoring and abandoned-cart recovery for upload journeys. Shipped as part of a growth initiative projected to drive incremental quarterly revenue.
Inspects controller callback chains at build time to flag endpoints missing authentication before they ship — surfaced and secured a previously-unprotected admin GraphQL controller.
Pushes post-shipment tracking links into Shopify for automated chargeback evidence, plus an HMAC-verified disputes webhook delivering real-time Slack alerts for dispute visibility.
Stripe webhook retries were creating duplicate sensor shipments. Implemented idempotency via event-ID deduplication and a DB unique index, with billing-period guards before shipment creation.
Designed a SQL-as-queue (a RabbitMQ replacement) for rate-limit-aware consumption of image/video generation across 15 model providers with limits ranging from 4 RPM to 150 RPS — covering task prioritization, exponential-backoff retries, and recovery for stuck jobs.
Conversational image/video generation agents built with Google ADK — a natural-language alternative to the platform's detailed configuration controls, lifting usage by 20%.
Refactored the GenAI tools codebase into an extensible provider architecture and integrated multiple image/video models across 4 request types.
Built backend APIs to generate and publish student assignment reports, integrating an OpenAI Assistant for automated generation and context management.
Re-architected the backend for multi-tenancy using School ID as the tenant key, leveraging async local storage for request-scoped context and strict data isolation.
Removed the default JPA findAll count query and fetch one extra record to detect the next page — eliminating a full-table count on every paginated request.
Refactored the notification service auth module with an interceptor and custom validators to handle all authentication request types through a single path.
Migrated order-report generation off BigQuery to self-hosted MongoDB, re-architecting the data layer in a Go + Gin backend while holding the required scalability and performance.
Order-notification events were firing on a JPA @PostUpdate callback — before the transaction committed — so a rollback could emit an orphaned event. Re-published them through a single async Spring @TransactionalEventListener that fires only in the after-commit phase, so an order is notified exactly when its shipment reaches a terminal state and the change is durably persisted.
Built segment-based discounts, pickup-point discounts, platform fees, shipment-level delivery charges, and a cash-reconciliation system tracking driver collections through to settlement.
$ ls projects
Built an application that detects fake news in text messages and links using a Random Forest model, returning a confidence score for each input.
$ ls education
B.Tech · Computer Science & Engineering