Supreet P

About

Software engineer with JPMorgan Chase experience delivering end-to-end fintech features (Next.js/React/Node/Postgres on AWS) and measurable performance gains. Built and productionized an AI-native credit decisioning workflow combining LLMs, vector retrieval, and a rules engine with strong governance (bias checks, auditability, human-in-loop), improving precision and cutting underwriting turnaround time by 40%.

Key Strengths

• End-to-end ownership across frontend, backend, data, and AWS infrastructure
• Shipped investor reporting/cashflow module with 35% report generation latency reduction
• GraphQL performance optimization for large datasets (batching/pagination/caching) reducing payload size by 40%
• Production monitoring and observability with CloudWatch and Datadog
• Built and productionized AI-native credit decisioning with LLM + rules engine + vector retrieval; 15% precision improvement and 40% faster underwriting turnaround
• Responsible AI implementation: anonymization/encryption (AWS KMS), bias checks, audit logging, human-in-the-loop for high-risk cases
• Legacy modernization leadership: monolith to modular React/Node with GraphQL compatibility layer, CI regression tests, feature flags; 25% stability improvement
• Strong stakeholder alignment and execution cadence (weekly syncs, Azure DevOps dashboards, rollout/rollback planning)
• Effective use of AI coding tools for architecture and debugging, cutting initial development time by 30%
• Designed usage-based billing + customer API with real-time metering and reliable invoicing
• Modular monolith architecture to preserve single transactional boundary (Postgres ACID) and reduce ops overhead
• High-throughput background processing with Celery, webhook/email retries, and Redis rate limiting
• Scaled large Postgres workloads (2B-row events table) using monthly range partitioning, targeted indexing (composite/partial/GIN), and SQLAlchemy Core hot paths
• Performance tuning with EXPLAIN ANALYZE, read replicas, PgBouncer pooling, and Redis caching to handle ~10x traffic
• Production ML platform integration: Airflow + Spark training, MLflow tracking, Feast feature store, ONNX model serving via FastAPI
• Handled data drift and low-latency serving with checkpointed Spark jobs, retries, versioned features/schemas, and bad-record isolation (dead-letter handling)
• Multi-tenant isolation across Kubernetes, S3 (KMS/IAM), and Postgres (schemas + RLS) with audit/monitoring
• Resolved Postgres RLS pitfalls (policy bloat, non-sargable filters, role leakage) using tenant_id checks, composite indexes, SECURITY DEFINER views, and forced row_security
• Kafka + Redis + batching + lag-based autoscaling improved write throughput from 30k/min (p95 1.8s) to 120k/min sustained (p95 120ms) and reduced DB load ~10x
• Delivered ~180ms P99 latency and ~99% uptime with zero-downtime blue-green deployments

Experience

Education

Skills

Languages