krish gupta gupta

Java Backend Engineer

Inspiron Labs Software Systems Private Limited

Oct 2024 - Present

HTG-CONNECT | Healthcare Provider Data Migration: Event-driven migration of provider data from legacy systems into DYP. Multi-GB TXT files land on Amazon S3, then a fail-closed 3-stage pipeline runs data sync, CFL conversion, and inbound MongoDB load for the UI. Designed a fail-closed 3-stage orchestration pipeline so 1 failed stage blocks the next 2 stages and bad data never reaches MongoDB or the UI. Raised ingest throughput from 1 sequential file read to 8-10 concurrent jobs with batch retries, using S3 pre-signed URLs for multi-GB TXT files and archiving each file only after all batches completed. Enforced tenant isolation during CFL conversion with per-tenant mapping rules, then started inbound only after conversion finished for that tenant. Owned inbound business logic across 5 microservices and 4 sub-modules. Used the Strategy pattern so new process modules can be added without nested conditionals. Reduced MongoDB write latency by replacing 1-by-1 MongoRepository iterable saves with unordered bulkWrite and 4 parallel sub-batches via CompletableFuture.runAsync(). Sustained 1M+ provider affiliation throughput with pagination, MongoDB indexes, and a shared-object cache across batches, which reduced Java heap pressure and out-of-memory failures. Horizontally scaled Kafka consumers so partition count matches replica count, increasing parallel throughput. Tracked CPU and memory in Grafana and Prometheus to reduce out-of-memory incidents. Used MongoDB Atlas replicas and Amazon RDS for PostgreSQL. AI-ROS | AI-Driven Restaurant Operating System: Multi-tenant, event-driven restaurant platform covering 4 areas: finance, budgeting, invoice processing, and operational data ingestion. Owned backend REST APIs for 4 workflows: data ingestion, budget synchronization, operational budgets, and financial processing across restaurant tenants. Designed a 4-stage S3-to-ClickHouse ingestion pipeline (staging, schema validation, general ledger mapping, SQS consumers) for operational uploads. Cut duplicate event risk with 3 reliability controls: idempotency keys, retries, and dead-letter queues on AWS SQS consumers. Synced budget versus actual figures across 4 invoice actions: commit, approval, reversal, and deletion, including point-of-sale and labor events. Applied JWT and restaurant-scoped authorization for tenant isolation. Tuned queries on 2 databases, ClickHouse and PostgreSQL, for ingestion and reporting latency.