Justin Velasco

Leveraged Google BigQuery and Gemini AI federated learning capabilities to integrate multiple data sources and improve RAG document retrieval and synthesis accuracy by 15%. Developed REST vector store services to abstract text embeddings from users. Accelerated ML model deployment by integrating Kubeflow for knowledge graph construction and document information retrieval. Integrated Gemini AI models into RAG workflows for advanced NLP tasks, improving response quality and contextual understanding in chatbots.

Designed and built conversational chatbots for healthcare and insurance, improving accuracy by 15% by integrating advanced LLMs and frameworks (React/Next.js, Django, Flask) with Pinecone vector databases. Engineered AWS SageMaker-based ML/IoT pipelines for health monitoring, improving storage efficiency in S3 and supporting incremental learning with Pandas/NumPy. Developed automated log analytics and repair using the Elastic stack combined with RAG, reducing resolution times by 40%. Led a team to build a B2B medicine supply chain portal with SQL/NoSQL data handling, achieving a 35% increase in operational efficiency. Built a fashion recommendations system that attracted 100,000 users in the first month using personalized recommendations. Developed a robust backend for an AI image generation platform (Flask, Docker, Jenkins, Runpod) and integrated with Dialogflow for dynamic, real-time user feedback.

Built a continuous learning ML system to serve and fine-tune models from user feedback. Integrated with Flowise and LabelStudio for streamlined user interactions. Engineered resilient data pipelines to collect and create training datasets, applied training techniques such as QLoRA, and implemented CI/CD using Kubeflow pipelines. Added Evidently-based model performance monitoring and improved inference speed by 20% using Google AI Platform services for training and deployment.