Rohit Singh

1. Led the development of an Intelligent Document Processor implementing Multimodal RAG architecture with LlamaParser and GPT4-0, leveraging LLM as Judge for quality control and validation. Achieved 95% accuracy in processing various financial documents through dual-LLM architecture where secondary LLM acted as judge for output validation. Implemented custom evaluation metrics for automated assessment of extracted data quality, reducing manual processing time by 70% while maintaining high accuracy across invoices, bank statements, and regulatory documents. 2. Leading the development of a flagship AI assistant, contributing to a differentiated product offering that empowers and streamlined core operations and enhanced client engagement in financial domain, contributing to a differentiated product offering that strengthened market position and supported the company’s growth. 3. Built a multi-modal data extraction system leveraging LlamaParser and a vision-based LLM to parse 50,000+ monthly scanned policies, handwritten claims, and PDF endorsements. The pipeline combined visual layout understanding (tables, stamps, signatures) with advance prompting to achieve 93% field-level accuracy for structured extraction of policy terms, coverage limits, and claim histories.

1. Designed and deployed a personalized hotel recommendation engine using AWS Personalize, integrating multi-source data (user clickstreams, interaction history, hotel descriptions, customer reviews, and Google Analytics user profiles). Trained a contextual recommendation model by combining behavioral, demographic, and content-based signals, then hosted the solution on AWS SageMaker for real-time inference. Automated daily batch predictions via cron jobs to trigger personalized trip suggestions and notifications for users’ upcoming travel plans, improving engagement by 30% and driving a 25% lift in booking intent across targeted campaigns. 2. Designed and deployed an advanced RAG system for personalized hotel discovery, integrating customer prompts with a semantic search engine trained on hotel embeddings (derived from descriptions, reviews, and amenities). The solution retrieves contextually relevant matches by comparing user queries against precomputed hotel embeddings stored in AWS OpenSearch, returning the top 5 recommendations ranked by similarity. Upon user selection, a conversational AI chatbot engages users in dynamic Q&A, providing granular details (e.g., pricing, policies, nearby attractions) and seamlessly redirecting to booking pages post-interaction.

1. Developed and deployed an attention-based LSTM model to optimize business development (BD) team strategies by analyzing historical conversational data between BD representatives and organizational leads. Trained on NLP-pre-processed transcripts, the model identifies patterns in successful negotiation tactics, objection handling, and lead conversion signals. The architecture integrates bidirectional LSTM layers with attention mechanisms to prioritize critical conversational segments (e.g., pricing discussions, timeline alignment) and predict optimal follow-up actions. 2. Developed and deployed a Personalized Customer Communication Engine leveraging a hybrid machine learning approach: first applying ensemble-based clustering to segment customer data, followed by classification models to predict best messaging channels. The solution was hosted on AWS SageMaker for scalable inference and configured with cron jobs to automate daily batch predictions. This end-to-end pipeline achieved 85% prediction accuracy and a balanced F1 score, driving measurable improvements in customer engagement and conversion rates across platforms including Epsilon, Attentive, and Snowflake.