ROHIT SINGH

Architected and deployed Intelligent Annuities (handwritten annuity application automation), reducing processing time from 4 hours to 8 minutes using page classification (CNN), vision-based OCR/markdown generation (Anthropic Claude) and schema translation (AWS Nova), with autonomous prompt optimization via LLM observability and continuous self-healing from corrections. Designed and implemented a Production Support agentic system to automate ticket triage, routing, resolution, and escalation using knowledge retrieval, contextual event memory, and HITL for novel issues. Built multimodal document processing with Multimodal RAG and LLM-as-judge validation; implemented evaluation metrics to improve extraction quality and reduce manual processing time. Architected a self-cure incident system for detection, root cause analysis, and automated remediation with code-aware indexing and retrieval.

Architected and deployed Intelligent Annuities, an end-to-end automated processing pipeline for handwritten annuity applications, reducing processing time from 4 hours to 8 minutes. This multi-stage AI system integrated a custom CNN for page classification, Anthropic Claude Opus 4.5 for vision-based OCR and markdown generation, and AWS Nova 2 Pro for intelligent schema translation. It features autonomous prompt optimization with Langfuse across a multi-LLM framework (Nova, DeepSeek) and a self-healing loop that learns from corrections to improve checkbox detection, handwritten number recognition, and form element extraction. Led Production Support Agentic System automating ticket triage, resolution, and routing via intelligent knowledge retrieval and autonomous decision-making, achieving a 70.2% automation rate. Developed an Intelligent Document Processor with Multimodal RAG (LlamaParser) using a dual-LLM setup with Gemini as judge for quality control, reaching 94.7% accuracy across fi

Designed and deployed a personalized hotel recommendation engine using AWS Personalize, integrating multi-source data (user clickstreams, interaction history, descriptions, reviews, and profiles) and hosted on AWS SageMaker for real-time inference. Automated daily batch predictions via cron jobs to deliver personalized trip suggestions, driving a 25.6% lift in booking intent. Designed and deployed an advanced RAG system for hotel discovery, combining consumer prompts with a semantic search engine trained on hotel embeddings stored in AWS OpenSearch, returning the top 5 matches and engaging users in a conversational Q&A that redirects to booking pages.