Abhishek Tiwari

AI-Powered Document Summarization for Visa Evaluation

Developed and deployed an intelligent document processing solution for Embassy operations, significantly streamlining the visa evaluation process. This AI-powered tool leverages open-source Large Language Models (QWEN, LLAMA) and Azure services to automatically extract and summarize critical information from visa application documents, including passport details, application forms, insurance, bookings, ITR, proof of occupation, and invitation letters.

Key Achievements:

• Reduced manual effort in reviewing extensive documentation, accelerating decision-making for visa grants.
• Implemented a robust summarization pipeline using FastAPI, deployed and hosted on Azure, ensuring scalability and reliability.

VISA AT YOUR DOORSTEP (VAYD)
Designed and implemented an AI-powered scheduling and route optimization solution using Python, Google Maps API, and Streamlit. Developed a multi-tab interface for Schedulers, Admins, and Agents to manage and track doorstep visa appointments in real time. Enhanced kit and field agent utilization, resulting in increased revenue and reduced operational overhead through efficient slot booking and intelligent routing.

Retrieval-Augmented Generation (RAG)

Architected and implemented a Retrieval-Augmented Generation (RAG) system leveraging OpenAI's Large Language Models, FAISS, and Chroma DB for efficient vector similarity search and storage. This solution enabled dynamic and accurate responses to user queries by retrieving relevant information from a dedicated knowledge base, significantly enhancing information retrieval and user satisfaction.

Led a 5-member engineering team in the successful development and deployment of an advanced AI-powered Text Insights Platform, capable of processing and analyzing textual data across six diverse languages.

This comprehensive solution leveraged cutting-edge machine learning and natural language processing techniques to create a suite of models for:

• Categorization: Identifying key themes (Agent, Process, Service, Technology).
• Predictive Analytics: Forecasting customer loyalty, churn risk, and journey stage.
• Interaction Analysis: Determining effort, contact intent, and specific issues.
• Affective Computing: Predicting sentiment and emotions from text data.

The platform was efficiently deployed on Azure Function Apps, ensuring scalability and real-time processing of large volumes of text, enabling data-driven strategic decisions and enhancing operational intelligence.

FCR PREDICTIVE MODEL

• Made a first call resolution model which predicts which customer will call back within 7 days to improve the first call resolution curve by 20%.