Arsheen Singh

1. Designed and developed the Agentic IT Solutions Chatbot to automate employee IT queries. It integrates context retrieval, memory, and action workflows through LangGraph. 2. Built the IT Hardware Agentic Chatbot in three versions: no-cost, low-cost, and advanced. This was done using LLM orchestration and Langfuse to evaluate cost efficiency. 3. Contributed to the PRD documentation, architecture design, and performance optimization for enterprise-grade AI assistants. This work improves internal IT service delivery.

Created TubeMate AI, a RAG-powered chatbot for YouTube videos via a Chrome extension and FastAPI service. Processed and indexed video transcripts using FAISS and MiniLM embeddings, delivering concise, context-aware answers through Llama 3. Achieved 92% answer accuracy across 500+ test videos and sustained 500 QPS under load testing. Containerized with Docker and orchestrated using GitHub Actions for CI/CD, enabling rapid deployment and rollback. Reduced viewer drop-off by 15% and increased video watch-time by 20% for users like The GrowWiser. Integrated user feedback loops and analytics dashboards (Streamlit) to monitor performance and guide iterative improvements. This project solidifies my expertise in real-time NLP applications, scalable inference, and full-stack AI deployment.

I completed my internship in Emerging Technologies (AI & Cloud). This 4-week program, offered by the Edunet Foundation in collaboration with AICTE, allowed me to develop my skills in key areas such as artificial intelligence and cloud computing. I gained hands-on experience using the IBM Cloud Platform and IBM SkillsBuild, which provided a strong foundation for real-world application.

·Engineered and deployed AI-powered chatbots and tools using Natural Language Processing (NLP) to automate customer support and streamline business processes. ·Managed the full lifecycle of AI applications, from developing user-facing interfaces and APIs to deploying them on AWS using Docker. ·Collaborated with cross-functional teams to translate business requirements into effective, user-centered AI solutions.

Delivered a data-driven analysis project by cleaning and transforming a 100,000-record dataset using Pandas. Carried out exploratory data analysis with Seaborn and Matplotlib to uncover key trends, leading to a 15% improvement in a classification model’s F1-score. Deployed an end-to-end inference pipeline using FastAPI, serving predictions in under 200ms. Presented findings and dashboards to stakeholders, driving recommendations that streamlined data collection procedures by 20%. Gained hands-on experience in model inferencing, version control (Git), and collaborative Jupyter workflows.

Completed a month-long AI/ML micro-internship, focusing on optimizing PyTorch models for GPU inference. Implemented Convolutional Neural Networks for image classification, reducing inference latency by 35% through quantization and batch normalization tuning. Learned transformer fundamentals by fine-tuning a BERT model for text classification with 88% accuracy. Collaborated with a mentor to document best practices in model versioning and code modularity. Contributed to team discussions on Gen AI applications, proposing a prototype AI agent for automated data annotation. This role sharpened my ability to translate research into practical, GPU-accelerated solutions.