Rushikesh Shaharao

Engineered and implemented security testing protocols for the Gen AI Security Platform, reducing LLM and chatbot vulnerabilities by 30%. Automated security assessment procedures, cutting manual effort by 40% and reducing turnaround time by 50%. Collaborated with cross-functional teams to refine testing methodologies, enhancing overall effectiveness and efficiency.

Led research on interpretability of BERT-like large language models (LLMs), emphasizing basic language features in text representation. Developed "Fictometer" algorithm achieving 90% accuracy in classifying text as fiction/non-fiction. Enhanced text classification accuracy to 98% by fine-tuning BERT models. Conducted experiments showing a single base model, fine-tuned for POS tagging and fiction/non-fiction classification, achieved high accuracy in both tasks.

Gathered and formatted diverse client data for RAG (Retrieval Augmentation Generation) Processes, incorporating BERT models for efficient embeddings. Spearheaded the execution of RAG Processes utilizing Vector DBs and MongoDB, deploying Q&A systems on WhatsApp and Telegram, reaching 2000+ users. Acquired expertise in fine-tuning Language Models (LLMs) and BERT models, optimizing natural language processing capabilities.

Developed a video platform feature using semantic similarity and embeddings. Enabled users to ask career-related questions and play videos from exact timestamps, improving content accessibility. Implemented cosine similarity for accurate query-to-video matching, guaranteeing a 95% accuracy rate for enhanced relevance. Conducted thorough testing and added an audio input feature for enhanced usability.

Systematically examined 100+ research papers, distilling key insights to address intricate ML and NLP challenges. Managed image annotation tasks using tools like LabelImg and Labelbox for CV projects. Applied research insights to contribute to practical solutions, advancing ML and CV projects.