Prajwal Deshpande

I built a multimodal RAG system that answers questions about PDFs using text and images. It generates cited answers, and my 24-question evaluation achieved a 91.7% top-3 retrieval hit rate, with expected evidence in the retrieved context for all 24 questions.

For my Crop Disease Prediction System, I trained a MobileNetV2 classifier on the PlantDoc dataset, reaching 95.05% test accuracy. I deployed it behind a Flask API and React frontend so users can upload a leaf image and receive a prediction.

As an open-source contributor to Kestra, I implemented cancellation support for Azure AI Foundry agent runs and added regression tests. I also developed a Neo4j polling trigger with integration tests; both pull requests were merged.

I'm a Computer Engineering graduate building LLM-powered applications in Python, with hands-on experience in RAG, LLM APIs, embeddings, and evaluation. I'm seeking an AI Intern role to build, test, and evaluate LLM features.