Vivek Kumar Rawat

Architected and shipped a production Generative AI document-extraction pipeline (PaddleOCR + AWS Textract + self-hosted multimodal LLMs) achieving 95%+ accuracy at ~50% cost reduction vs vendor solutions — full ownership from design through deployment
Built Python (FastAPI) + PostgreSQL backends with async data pipelines, WebSocket streaming endpoints, and structured logging for production reliability
Developed real-time dashboards in React/TypeScript consuming live WebSocket feeds — sub-second latency on streaming data
Engineered interactive e-commerce platforms in Next.js with optimised data fetching, improving user engagement and conversion metrics
Adopted AI-native development workflows (autonomous coding agents, automated test generation) to accelerate delivery velocity
Co-founded a 3-person startup: contributed to product architecture, equity structuring, and early-stage business development

Client: Yamaha (E-commerce, US)
Built AI-driven product discovery modules with NLP-based filtering, improving search relevance and conversion rates
Optimised real-time data ingestion pipelines for inventory and catalogue sync between backend services and frontend
Developed data visualisation dashboards in Next.js surfacing user-behaviour insights for marketing and data teams
Built automated A/B testing frameworks to measure recommendation engine impact on engagement metrics

Client: Natixis (Investment Banking)
Built financial intelligence dashboards using ECharts and AntV for predictive analytics and risk visualisation
Implemented real-time data streaming with GraphQL Subscriptions and schema stitching for high-frequency financial feeds
Reduced rendering latency by 40% for large-scale, data-heavy financial modelling tools through virtualisation and memoisation
Built a reusable component library for high-density data UIs deployed across multiple banking applications

Client: Credit Suisse (Investment Banking)
Built decision-support interfaces for trading platforms integrating real-time ML-driven signals with sub-100ms display latency
Developed high-throughput data layers for low-latency rendering of algorithmic trading outputs
Designed data validation and error-handling protocols ensuring integrity in risk-modelling workflows
Led legacy-to-React migration for core trading UI modules, reducing codebase complexity by 35%

Client: Capital One (Banking)
Automated document verification workflows using OCR, reducing manual data-entry time significantly
Built intelligent form-validation engines using pattern recognition to catch application inconsistencies upstream
Engineered scalable, high-accuracy data-entry interfaces feeding downstream ML and audit pipelines
Delivered cross-platform reporting modules for consistent data visualisation across departments