r
ramesh_36

ramesh_36

Índia
Telugu
Algumas informações são exibidas no idioma inglês.
Sobre mim
I worked as a Data Scientist building end-to-end AI systems across CV, OCR, NLP, Fraud Detection, and GenAI. I developed a KYC automation pipeline using FastAPI, YOLOv5, and PaddleOCR, reducing processing time by 42% and handling 3+ crore documents monthly, saving ₹14 crore. I also built a face deduplication system using DeepFace and ChromaDB for fraud detection, and a RAG-based chatbot using Azure OpenAI and FAISS, improving internal efficiency by 60% with fast policy query resolution.... Saiba mais

Habilidades

r
ramesh_36
ramesh_36
offline • 

Conheça meus serviços

Visão Computacional
I will build ai ml solutions for ocr nlp computer vision and genai
A partir deUS$ 90 /projetoMais detalhes

Experiência profissional

Data Scientist

TVS Credit • Período integral

Aug 2024 - Present • 1 yr 9 mos

I worked as a Data Scientist responsible for architecting and deploying end-to-end AI systems at enterprise scale across Computer Vision, OCR, NLP, Fraud Detection, and Generative AI. I led the development of a production-grade KYC document automation pipeline that replaced manual verification for millions of customers. This solution was built using modular FastAPI microservices integrated with YOLOv5 for robust document classification and PaddleOCR for high-accuracy text extraction. I engineered GPU-optimized inference workflows using NVIDIA A10 GPUs and pynvml for intelligent resource scheduling, reducing average processing time from 12 seconds to 7 seconds (a 42% improvement). Deployed on Oracle Cloud Infrastructure with autoscaling and load balancing, the system processes over 3 crore documents per month and has handled more than 45 crore documents to date, delivering an estimated 14 crore rupees in operational cost savings. The pipeline seamlessly integrates with enterprise onboarding systems via REST APIs, providing high reliability, faster turnaround time, and over 30 hours of weekly manual effort eliminated. In parallel, I developed a real-time Face Deduplication and Fraud Detection system to help investigators identify duplicate or fake identities across loan databases. This involved creating a secure AI web application that allowed authenticated teams to enter loan or customer IDs and instantly retrieve all associated KYC records. I implemented a deduplication workflow using DeepFace and stored 16+ lakh face embeddings in ChromaDB for efficient similarity search. The system returns top matching faces with metadata, enabling investigators to detect suspicious overlaps within seconds. I also designed an automated discrepancy-detection mechanism that cross-checks customer information across records and flags inconsistencies to the Risk Control Unit (RCU), significantly strengthening fraud-prevention capabilities across the organization.