Experience

Key Responsibilities

• Architected a distributed reconciliation system to process 2M+ daily financial transactions, utilizing Kafka with custom key-based partitioning to ensure sequential processing by merchant.
• Engineered fault-tolerant data ingestion with idempotent AWS Lambda consumers, implementing exponential backoff retries and a Dead-Letter Queue (DLQ) for robust poison message handling.
• Optimized data storage and analytics using S3 Parquet columnar format with date-hour partitioning, reducing analytical query latency from 45s to 2s and enabling real-time financial reporting across multiple data centers.
• Designed enterprise RBAC system using OPA/Rego for microservices authorization, implementing 200+ permission policies across 15 services with centralized distribution via Redis cache and auto validation pipeline, reducing authorization latency from 50ms to 5ms.
• Built end-to-end Payment AI Agent using LLM APIs with React/TypeScript frontend and Python FastAPI backend, implementing real-time screenshot analysis, contextual invoice generation with validation pipelines, and A/B testing framework via Amplitude.
• Optimized high-traffic payment UI architecture using React virtualization with react-window, implementing progressive loading for 100K+ transaction records, lazy component hydration, and comprehensive E2E testing via Playwright reducing page load time by 60%.

Technologies Used

Key Achievements

• Processing 2M+ daily financial transactions
• Reduced query latency from 45s to 2s
• Reduced authorization latency from 50ms to 5ms
• Reduced page load time by 60%

Key Responsibilities

• Led development of a large-scale social media intelligence platform to identify and track misinformation campaigns, processing 100K+ daily records using Selenium/Python, Redis, and Docker ML orchestration.
• Pioneered novel Belief Embedding Model integrating BERT, VGAE, and Segment Trees for misinformation detection, achieving breakthrough performance on social network analysis benchmarks.
• Built React-based analytics platform with 3D/2D network visualizations serving 50+ researchers, implementing automated evaluation frameworks and enabling 15+ published papers across multiple institutions.
• Designed production-grade ML infrastructure with continuous retraining, A/B testing, and real-time classification workflows, supporting research collaborations and cross-platform misinformation analysis.

Technologies Used

Key Achievements

• Processing 100K+ daily records
• Breakthrough performance on ML benchmarks
• Serving 50+ researchers
• Enabled 15+ published papers

Key Responsibilities

• Developed React-based deposit list and tracker with advanced data filtering, sorting, and CSV download functionality using MySQL backend, processing 5K+ daily deposit records and improving user satisfaction scores by 15%.
• Implemented deposit management feature using Spring Boot with RESTful APIs for status updates, added input validation and auto email notifications for status changes, reducing failure rates by 25% and improving search performance through database indexing.

Technologies Used

Key Achievements

• Processing 5K+ daily deposit records
• Improved user satisfaction by 15%
• Reduced failure rates by 25%

Key Responsibilities

• Implemented Redis caching layer for product catalog queries to reduce DynamoDB load from 8K+ QPS, using TTL-based expiration and achieving 92% cache hit ratio, reducing search latency from 300ms to 80ms.
• Built cache invalidation system using DynamoDB Streams and Java to automatically evict stale product data when catalog updates occur, maintaining cache consistency across frequent product changes.
• Developed CloudWatch monitoring dashboard to track cache performance metrics and automated alerts for cache degradation, improving incident response efficiency.

Technologies Used

Key Achievements

• Reduced DynamoDB load from 8K+ QPS
• Achieved 92% cache hit ratio
• Reduced search latency from 300ms to 80ms