Developing AI-powered video generation platform with advanced customization features.
Key Achievements
Introduced advanced image customization utilizing fast-trained LoRA Built Chat-to-Video service leveraging Vercel AI and Langchain agents Migrated LLM integration from Azure OpenAI to AWS Bedrock Managed multi-environment infrastructure on AWS, managing instances of 4 independent microservices Containerized applications using Docker for consistent builds and deployments across environments Refactored Next.js frontend for improved UI/UX using Headless UI and ShadCN UI Technologies
Next.jsNode.jsAWSDockerGCPMySQLVercel AILangchain
Developed automated business solutions and tax optimization features.
Key Achievements
Developed automated business budgeting feature using Langchain Agent Improved product utility with personal tax optimization feature Identified tax deductible transactions for enhanced user engagement Implemented tRPC API for seamless frontend-backend communication Technologies
Next.jsLangchaintRPCSupabasePostgreSQL
Advanced research on Transformer model efficiency and context length optimization.
Key Achievements
Implemented methodologies to enhance Transformer model efficiency Extended context length capabilities of AI models Organized meetings to integrate diverse architectural frameworks Tested RetNet, VQ-VAE, and RoPE frameworks for model improvement Technologies
PythonPyTorchTransformers
Research on music generation with fine-level dynamics and mood classification.
Key Achievements
Implemented fine-level dynamics in music generation tasks Achieved 80% success rate on valence mood classification using SVM Conducted exploratory data analysis on music datasets Research accepted to AAAI 2023 conference Technologies
PythonScikit-learnSupport Vector MachineExploratory Data Analysis
Spearheaded research on combinatorial music generation published in NeurIPS 2022.
Key Achievements
Led ablation study for NeurIPS 2022 paper (ComMU) Investigated metadata effects in combinatorial music generation Evaluated distinctiveness through Amazon MTurk with 100+ composers Developed internal knowledge sharing templates for organization Technologies
PythonMachine LearningAmazon MTurk