At Carer Healthcare AI, I designed and implemented a production-grade evaluation framework for conversational Voice AI systems. I also built an end-to-end Voice-to-Voice evaluation pipeline that works in the audio embedding/token domain without an ASR dependency.
At Carer Healthcare AI, I worked on implementing and training a diffusion-based model for a Voice2Voice task with no text involved.
I fine-tuned a T5-small Transformer with LoRA (PEFT) for text summarization, with 0.96% of its parameters trainable. I also deployed a responsive Streamlit app for real-time summarization with adjustable output length.
I developed PDF Chat, a RAG-powered document assistant using LangChain and Groq API, and built AutoML WebLab, a Django platform for training and evaluating machine learning models.
