Sakshi Rathi
AI Engineer · LLMs, RAG & Fintech AI Systems · MS Data Science, Indiana University
AI Projects
LLMs & Fine-Tuning
| BERT Fine-Tuning for Phishing URL Classification | Python, BERT, Hugging Face, PyTorch |
Fine-tuned BERT (bert-base-uncased) for binary phishing URL classification using Hugging Face Transformers. Achieved 90% accuracy and high AUC on a real-world threat dataset, outperforming baseline ML classifiers. Built the end-to-end pipeline including tokenization, custom classification head, and hyperparameter tuning — demonstrating production-ready NLP fine-tuning practices.
| Sentiment Analysis with BERT + LoRA | Python, LoRA, Hugging Face |
Fine-tuned a pre-trained BERT model for sentiment classification using Low-Rank Adaptation (LoRA) and the Hugging Face ecosystem. Classifies IMDB movie reviews as positive or negative, achieving 90% accuracy on the validation dataset.
RAGs
| RAG-Based Chatbot | Python, LangChain, FAISS, Gemini, Streamlit |
RAG-powered chatbot with a unified Streamlit interface to query both PDF-based EV reports and CSV sales data. Uses Hugging Face embeddings and FAISS vector search for document retrieval, and Gemini Pro via LangChain for accurate, source-grounded responses. Also includes an automated SQL-generation pipeline using LangChain’s SQL Agent with Gemini Flash — enabling non-technical users to query structured data conversationally.