Ranip Hore

Designed and deployed end-to-end AI systems for customer support, including machine learning models, backend APIs (Flask/FastAPI), and Kubernetes-based deployments. Built an enterprise-grade conversational AI system with FastAPI, leveraging LLaMA-based RAG, tool-calling framework, and SSE streaming. Implemented scalable conversation state management with Redis + DB persistence, modular tool registry, debug tracing, and adaptive workflows (clarification, options, artifact generation). Developed Retrieval-Augmented Generation (RAG)-based question-answering system using hybrid search (keyword/semantic), query reformulation, cross-encoder re-ranking; integrated with Oracle documents, knowledge materials and Confluence for faster issue resolution, cutting support search time by 40%. Built Similarity Recommendation Tool for service requests with hybrid search models, recommending related requests to accelerate solutions; deployed in Kubernetes for scalability. Created LLM-based Clustering Framework for service requests, enabling dynamic incremental clustering; filed patent and reduced manual triage by 30%. Conducted statistical analysis on clusters to identify trends and hotspots, enhancing prioritization efficiency, and improved resource allocation by 20-30%.

Designed DIME data processing engine for investment banking client, normalizing data from multiple sources (various formats) for trading decisions, integrated third-party providers. Developed PyFTP Python package using SFTP/LFTP for downloading trading data from remote servers, streamlining data ingestion. Containerized applications with Docker and deployed to Amazon EKS (Kubernetes) clusters for scalable ML pipelines.

Automated financial processes using AWS Textract for SAP invoice processing, generating expense reports and saving man-hours. Built RPA workflows for daily BI reports, optimizing accuracy for finance leadership. Automated financial restatement processes to ensure timeliness and precision.