SURENDRA KUMAR

Developed an intelligent document extraction system for parcel images, extracting logistics data (dimensions, weight, AWB) with 98%+ accuracy. Cut AWS costs by 70-80% via dynamic EC2 auto-scaling and GPU lifecycle management. Integrated YOLO-based anomaly detection with ensemble OCR (PaddleOCR + EasyOCR). Built a distributed Celery/Redis pipeline for concurrent batch processing with real-time tracking and deployed production-ready infrastructure using Nginx, SSL/TLS, and systemd. Built end-to-end agentic AI workflows for automated lead enrichment, qualification, and multi-channel outreach (Email, WhatsApp, LinkedIn), integrated with HubSpot CRM, improving lead throughput and engagement tracking. Designed an intelligent vendor compliance pipeline that extracts and validates ESIC/EPFO data from PDFs using pdfplumber, tabula and PyMuPDF, allowing automated cross-verification of workforce submissions and proactive penalty detection. Built a high-performance P&L automation engine in Python that reduced daily financial processing time from 34 hours to under 3 minutes by implementing zone-based rate intelligence, dynamic surcharge calculations and multi-courier analytics with automated MIS reporting. Designed and deployed ProCompliance, a compliance task management platform with multi-tier RBAC (SuperAdmin/Admin/Viewer), domain-restricted Google SSO, automated stakeholder reminders, and flexible S3/filesystem storage, improving compliance visibility and accountability.

Leveraged advanced GenAI models (LLMs & VLLMs) to build intelligent FMCG retail systems for product pricing, promo detection, and data tagging automation. Trained and optimized ResNet, MMDetection, and Detectron2 models for classification, detection, and segmentation. Engineered OSA & SOS analytics pipelines to monitor product availability and analyze shelf share, enhancing retail insights. Streamlined AI workflows employing Docker, facilitating scalable model deployment and operational efficiency.

Constructed an AI-powered DeFi aggregator on the Aptos blockchain to enhance global accessibility and user experience. Fine-tuned LLMs to enable swift prompt-driven DeFi transactions, achieving a 30% improvement in processing speed. Developed autonomous agents using LangGraph and LangChain, orchestrating real-time API interactions and facilitating intelligent decision-making. Designed a retrieval-augmented generation (RAG) pipeline that improved DeFi information retrieval efficiency by 25%.

Created and deployed LLM-powered chatbots, RAG pipelines, and diffusion models for enhanced text and image generation, driving a 40% surge in user interaction. Developed real-time APIs utilizing FastAPI, Flask, and WebSockets, focusing on NLP for automated SQL query generation. Engineered a real-time document verification system using computer vision and OCR, reducing verification time by 30%. Built an Applicant Tracking System, optimizing the hiring process by 20% and improving candidate screening efficiency.