Fine-tuning state-of-the-art "smaller" language models for a text-classification task to replace human labeling. Targeting 99% accuracy or higher. Expected to be in production in Q3 2024 (Python, PyTorch, Hugging Face, PEFT, Accelerate, DeepSpeed, and LoRA). - Fine tuning different pre-trained models to determine the accuracy they can achieve. - Created a distributed multi-machine multi-GPU training environment in AWS. Developed LLM generated descriptions of cities and neighborhoods from proprietary content. Launched in July, 2024 on Trulia brand. (Python, LangChain, OpenAI, and gpt-4-32k).- Developed LLM prompts to generate the descriptions, in a very specific format, containing a headline, likes and dislikes. - Designed LLM prompts to evaluate compliance with the Fair Housing Act (FHA) and modify any non-compliant text into compliant forms.Competed in the Kaggle Zestimate (Zillow) competition as an individual, where I was not eligible for a prize. Achieved bronze medal (Python and Keras).Created an SEO crawl monitor by analyzing bot crawl logs and generating insights (Java, Apache Flink, Python, AWS Lambda, PostgreSQL, and complex SQL queries).Constructed data pipelines to efficiently organize imported data, enabling it to be served through microservices (Apache Spark, Scala, AWS Glue, PostgreSQL, and stored procedures).Developed numerous microservice to serve data to users (Java, Spring Boot, TypeScript, NestJs, MySQL, and PostgreSQL).Developed multiple serverless applications for periodic and event-driven tasks (AWS, Terraform, and Python).Created services and systems infrastructure in cloud (AWS, and Terraform).