Vishal Parekh

🔹Leading the modernization of internal AI platforms by migrating legacy Prompt Flow based pipelines to a LangGraph-driven RAG architecture, improving orchestration, observability, and extensibility across use cases. 🔹Actively developing and scaling an in-house AI chatbot for customer support, serving multiple banking products including Wealth Management and Credit Cards, with secure, compliant, and production-ready LLM integrations. 🔹Leveraging an ingestion framework to efficiently index enterprise data into Azure AI Search, enabling high-quality retrieval through prompt-optimized indices. 🔹Designed and implemented automated ingestion pipelines to convert SharePoint sites into structured knowledge-base indices, supporting continuous knowledge updates for RAG workflows. 🔹Configured and managed cron-based ingestion jobs to keep search indices and model contexts up to date with minimal operational overhead. 🔹Collaborated closely with platform, data, and security teams to align AI solutions with enterprise governance, compliance, and reliability standards.

🔹Designing and implementing a multi-agent AI system to automate ticket prioritization and retrieve customer health data using sentiment analysis and contextual metadata. 🔹Developed a RASA-based technical support chatbot that autonomously resolves 85% of user queries by leveraging custom conversational flows and optimized intent classification. 🔹Led the design and deployment of an Email Folder Suggestion System using semantic similarity and Pinecone, enabling streamlined email classification for improved user efficiency. 🔹Built and launched an RAG-based FAQ system with LangChain, ChromaDB, OpenAI GPT-4, and Streamlit, deployed on Hugging Face Spaces to provide dynamic, context-aware answers. 🔹Developed a multilingual RAG pipeline for a law firm using LlamaIndex and PgVector, incorporating dynamic prompt templates to handle language-specific nuances in document retrieval. 🔹Built Python automation scripts to detect and troubleshoot system errors, reducing manual intervention by 60% and improving response times. 🔹Fine-tuned Meta’s Llama 3 using LlamaIndex and Hugging Face on proprietary datasets for specialized tasks like entity recognition and classification. 🔹Deployed optimized models into production using quantization and pruning techniques, reducing inference latency by 25%. 🔹Designed and implemented NLP preprocessing pipelines for metadata extraction, significantly enhancing accuracy in document search and retrieval.

🔹 Orchestrated the migration of a legacy EPR platform to a Node.js-based system, reducing server response times by 50% and supporting a 3x increase in user transactions. 🔹 Collaborated with cross-functional teams, integrating Node.js services into a microservices architecture, resulting in a 20% increase in service scalability using AWS ECS. 🔹 Developed and maintained front-end interfaces using React, enhancing user experience with dynamic content rendering and modular component architecture. 🔹 Integrated GraphQL into existing RESTful services, enhancing data retrieval flexibility for front-end developers and reducing over-fetching issues by 30%. 🔹 Automated deployment processes using Docker, decreasing deployment time by 40% and enabling CI/CD workflows with Jenkins.