-Stable Diffusion models•Build a pipeline of stable diffusion models using open sourced checkpoints and Lora adapters•Deployed serverless pipeline for text2image generation for integration in apps-RAG based application•Integrated Mongo vector db for storing text embedding for efficient RAG pipline•Integrated a database (Milvus vector db) for storing text embedding for efficient RAG pipline•Conducted iterative experiments for query search with search metric and indexing algorithms like "L2","cosine" with "HNSW", "IVF_FLAT"-Talking Head• Led a project creating talking heads from still images using a SadTalker model that processes audio andimages.•Conducted in-depth research, extensively reviewing numerous papers and code repositories, to identifya state-of-the-art model.•Conducted iterative experiments to optimize hyperparameters, ensuring the model's performancealigns with project objectives.•Integrated OpenAI's TTS for text-to-audio conversion, connecting it seamlessly with the model for videogeneration.•Deployed model on serverless serving using runpod services-Integration of open source LLM•Evaluated model outputs, considering the training process and techniques such as SFT, RLHF, and DPO.•Utilized libraries such as Hugging Face to implement and streamline the evaluation of open sourcelanguage models (LLMs) of different sizes and hyperparameters to fit the use case.•Conducted analysis to generate structured output through LLMs.•Performed benchmarking on LLMs, assessing token counts and inference speed.•Conducted benchmark tests on diverse hardware platforms for comprehensive evaluation.•Contributed to prompt engineering to enhance model performance and responsiveness.

Worked on training GANs for Super Resolution (Domain adaptation)