Vivek Kumar

About Vivek Kumar

AI & Software Engineer with 2+ years of experience building scalable, production-ready Generative AI (GenAI)and Natural Language Processing (NLP) solutions on Microsoft Azure. Hands-on experience implementing and enhancing RAG and document intelligence workflows using Azure OpenAI, LangChain, and Python to automate enterprise claim processing and support high-volume workloads. Skilled in processing large-scale unstructured data and delivering reliable cloud-native AI solutions for enterprise automation.Experienced in optimizing AI system reliability, cost efficiency, and responsible AI practices. Adept at building robust, efficient, and compliant AI systems for real-world enterprise applications.

Experience

1. System Engineer

   Tata Consultancy Services

   Dec 2023 — Present · Hyderabad, IN

   Project: Medical Claim Documents Summarization | Azure AI & Generative AI• Built an end-to-end medical claim document summarization and classification pipeline using Azure AI Document Intelligence (Read & Layout OCR) and GPT-4.• Automated processing of PDF and ZIP-based medical documents in Azure Cloud, including PDF-to-image conversion and image extraction, reducing manual effort by ~40%.• Generated page-level structured text outputs while preserving layout, tables, and spatial context for downstream NLP workflows.• Implemented a hybrid classification workflow combining rule-based logic and LLM-driven (GPT-4) classification, efficiently handling 30+ page multi-class and 50+ page single-class documents.• Extracted key dates and metadata using prompt engineering, enabling temporal grouping, chronological ordering, and delivery of automated, class-wise, date-sorted outputs for medical claims analytics.Project: Claims Call Transcript Summarization | Generative AI• Built a GenAI system using Azure OpenAI (GPT-4) to analyze and summarize phone call transcripts between claims agents and customers, extracting structured insights.• Scaled to process calls daily, reducing manual review effort by ~80%.• Led domain-specific prompt engineering with role-based instructions, few-shot examples, and strict output schemas, achieving 90%+ extraction accuracy while reducing token usage by ~40%.• As part of the broader solution, the system incorporated LLM evaluation using the DeepEval framework to assess extraction quality and consistency.• Implemented PII detection and masking using Azure AI Language – Text Analytics API (PII Recognition) to ensure data privacy and regulatory compliance.

Education

• Bharati Vidyapeeth's College of Engineering, New Delhi

  Bachelor of Technology , Computer Science and Engineering