• Implement AI-driven features using GPT-4o, Gemini 1.5, Claude-3.5 Sonnet, and LLama models, improving existing AI capabilities and end-user experience.• Develop production-grade RAG systems, LLM Solutions, and apply agentic patterns to create AI-powered assistants, improving product engagement by 35%.• Design scalable AI architectures leveraging distributed computing, cloud services (Azure, AWS, GCP)• Prototype and test new AI ideas, rapidly validating hypotheses and aligning with business goals, resulting in 3 successful product launches.SELECTED PROJECTS--- GenAI for Enterprise• Developed a Generative AI chatbot using Microsoft 365 and other enterprise data sources, applying agentic models to create a seamless user experience.• Integrated AI-powered data pipelines, automating workflows and reducing manual interventions across departments by 50%.--- Finance AI• Architected and implemented a generative AI-powered financial assistant using state-of-the-art LLMs (GPT-4, Claude-3.5 Sonnet), improving decision-making speed and accuracy by 40%.• Designed and developed effective prompts to interact with LLMs, refining them based on model responses and specific use cases to ensure engaging and accurate outputs for users.--- eDaters.gg• Design and launch a gamified online dating platform, using Python, reaching 1200 users within three weeks and generating $2000 in revenue in the first month.• Implemented AI-powered features to enhance user experience, using RAG systems and agentic patterns to create personalized recommendations and interactions.--- AI Game Assets• Fine-tuned Generative AI models for producing video game assets, automating creative asset generation, and improving product timelines by delivering key features using Node.js, Python, and React.• Developed and iterated on AI proofs-of-concept to test new ideas and hypotheses, efficiently validating and evolving AI initiatives for game asset generation.

• Consult and build Full Stack applications for clients in Enterprise, Finance, Sports, Marketing, and Dating domains.• Deploy applications using AWS services including Fargate, EC2, ECS, Cloudwatch, and more.• Design and implement custom and traditional RAG architectures, achieving 90% accuracy on evals.• Evaluate LLM models to balance budget, iteration speed, accuracy, and stakeholder expectations.• Architect and assemble a ”data-first” platform to service MLOps and Data Ops for modern AI/ML practices.• Reduce third party LLM APIs costs by 55% via data cleaning techniques and prompt engineering.• Transform prototypes to production, enhancing data ingestion speed by 80% and lowering storage costs by 15%.• Interview, onboard and lead tams of developers and a product manager, utilizing Linear app (JIRA alternative) for agile project management, including handling Sprints, User Stories, Use Cases, and Kanban visualizations• Launch multiple web applications reaching 1000 users in the first few weeks and 40% retention after the first week

- Analyze and develop predictive behavior models using sci-kit learn, PyTorch, and NumPy for video game user performance• Establish statistical models, achieving 85% accuracy for classifying and predicting user performance.• Enable players to assess individual and team performance, elevating player performance on all KPIs by 15%.• Apply modeling techniques such as SVM, KNN, Stochastic Gradient Descent, Random Forests

● Leveraged Data Science and AI expertise to develop a real-time Computer Vision tool for a competitive video game team, employing OpenCV, PyTorch, and Optical Character Recognition for data extraction and Time Series dataset creation.● Improved YoloV5 performance by implementing fine-tuning using TensorFlow on a custom dataset, elevating detection from 60% to 90% with 95% consistency for both images and text.● Enhanced data validation by optimizing OpenCV and TessOCR, achieving 99% accuracy and conversion, while significantly increasing screenshots per second from 2 to 15.● Automate post-game analysis, enabling real-time performance reviews and increasing team productivity by 70%.

• Process and label 10,000+ images to improve PyTorch image classification model accuracy from 60% to 98%.• Improved an image classification tool on a custom dataset using transfer learning in PyTorch.• Review academic research papers every week to identify relevant techniques in model architectures.• Research, implement, and improve Convolutional Neural Net models, going from 83% to 96% accuracy.• Serve as the leading technical expert, guiding a team of 7 biology and oceanography students in programming.