Arnav Saini

Generative AI Chatbot and Forecasting Modeling

• Developed conversational chatbots using OpenAI GPT-4o-mini and Google Gemini for personalized travel recommendations via text and voice.
• Implemented predictive forecasting models based on routes and historical data to proactively suggest relevant travel options.

Impact

• Reduced booking process time by approximately 60%.
• Increased the number of transactions by 23% due to enhanced user engagement.
• Significantly improved overall user experience through personalized and automated recommendations.

Policies Implementation using Agentic AI

• Automated airline policy implementation on nexusKite using Agentic AI (Gemini and LangGraph), enabling clients to directly specify desired policies via natural language queries.
• Developed AI-driven agents to automate complex airline policy workflows, significantly streamlining the policy creation and implementation process.

Impact

• Drastically reduced the new airline policy implementation time from 1 to 2 months to just 2 to 3 days.
• Enhanced efficiency and accuracy of airline policy management through the use of AI-driven agents.

Generative AI and Large Language Models (LLMs)

• Developed a fine-tuned Retrieval-Augmented Generation (RAG) system using LLMs like Llama-2 and Mistral to automate chatbot creation and SQL query generation, resulting in a 27% reduction in manual workload.
• Leveraged LLMs like Gemini Pro Vision, Gemini, and Gemma to extract actionable insights from datasets and charts, improving efficiency by 40%.

Natural Language Processing (NLP) and Natural Language Understanding (NLU)

• Fine-tuned BERT and GPT-2 models for sentiment and intent classification, enabling enhanced understanding of customer needs and improved customer experience metrics NPS of above 60%.

Time Series Forecasting and Demand Management

• Implemented time series forecasting models to predict customer demand with 93% accuracy, leading to optimized inventory management and reduced stockouts.

Price Elasticity Analysis

• Employed price elasticity analysis to optimize discount strategies across products and stores, achieving a 20% cost reduction while maintaining stable demand.

Propensity Modeling

• Leveraged supervised machine learning to develop a customer visit prediction model. This model forecasts customer behavior for the next month, enabling targeted marketing campaigns and improved offer personalization, which result in increased conversion rates by 7% (deployed on Microsoft Azure).

Customer Segmentation

• Implemented unsupervised machine learning for customer segmentation. This data-driven approach created distinct customer personas, leading to more effective marketing strategies, and improved customer targeting (utilized K-means and K-means++ clustering).

Recommendation Systems

• Designed a product recommendation system using matrix factorization techniques. This system personalizes product suggestions for each customer, enhancing customer engagement, and driving sales ("built a collaborative filtering recommender system using PySpark and the ALS method").

• Spearheaded the development and deployment of high-impact predictive models for Zomato restaurant ratings (link: https://github.com/pranaysawant/ZomatoRestaurant-Rate-Prediction) and building energy efficiency management systems (97% accuracy).
• Leveraged machine learning on AWS to predict restaurant ratings, potentially improving customer experience and informing business decisions.
• Designed and built 'Jarvis,' a voice recognition system for voice commands and home automation, showcasing cutting-edge NLP skills and a user-centric approach.
• Developed a Deep Learning solution using OpenCV to bridge the communication gap for hearing-impaired individuals by recognizing hand signs and translating them to text.
• Built a Phishing Website Detection model (95% accuracy) to protect users from cyber threats.
• Implemented an Automatic Vehicle Number Plate Detection System using OpenCV.
• Developed a Handwritten Digit Recognition model and an advanced recommendation system using content-based and collaborative filtering techniques.

• Gained practical experience in building and implementing algorithms from scratch through data science projects. This approach fostered a deep understanding of how algorithms function and their impact on real-world applications.
• Implemented NLP, LSTM, word-2-vec, Bag of words, for sentiment and NER understanding. Implemented Transformers from scratch using PyTorch.

• Company Overview: (TCS-iON contract)
• Ensured seamless network operations by managing Local Area Networks (LANs) and Wide Area Networks (WANs).
• Implemented efficient routing protocols and configured switching systems for optimal network traffic flow. (VLANs for switching).
• Safeguarded the network by administering firewall security systems. Proactively monitored network activity to mitigate potential security threats.
• Network devices: Routers, switches, firewalls.
• Network protocols: TCP/IP, HTTP, DNS.
• Network security concepts: VPN, encryption, intrusion detection/prevention system (IDS/IPS).
• (TCS-iON contract)