Abhinav Nagathan

Designed and developed scalable movie discovery platform utilizing TMDB API, showcasing over 1,000 titles with real-time filtering.

Implemented reusable components and hook-based architecture to enhance UI modularity and minimize re-renders.

Integrated pagination and genre-based filtering to reduce load times and improve user navigation.

Created Favorites feature with persistent local storage, enabling users to save preferences across sessions.

Incorporated fuzzy search and state management to enhance content discoverability and decrease average search time.

Deployed application on Vercel, ensuring responsive design across devices.

Achieved increased content discoverability, improved retention rates, and doubled user bookmarking efficiency.

Designed structured clinical prompts to facilitate step-by-step diagnostic reasoning.

Fine-tuned DeepSeek-R1-Distill-LLaMA-8B and LLaMA-3.2 Vision models on radiology QA datasets using LoRA adapters.

Employed Unsloth for efficient fine-tuning, optimizing GPU memory usage, and securely uploaded trained adapters to Hugging Face Hub.

Preprocessed radiology datasets and tokenized data with Hugging Face tools for model compatibility.

Developed a lightweight Streamlit web app enabling real-time symptom input and structured diagnostic suggestions.

Evaluated model performance on MedQuAD and PubMedQA datasets, focusing on accuracy and F1-score.

Assessed model responses through manual review for medical relevance and reasoning depth.

Explored follow-up pipeline options to address unclear symptom inputs for future enhancement.