GenAI Labs Team

Engineering an adaptive simulation system using Python and C# for backend, generating autonomous robotic simulations through optimized code synthesis and behavioral modeling under the supervision of EECS Postdoc Edward Kim and EECS Prof. Sanjit Seshia. A BAIR Open Research Commons project in partnership with Meta (FAIR)

🛠️

UC Berkeley’s first and largest Generative AI organization.

LLM routing system for efficient query handling

Worked on the development of an image generation platform which boosted sales and consistently surpassed delivery expectations on tight deadlines.Developed a transcription and summarization system for over a thousand hours of videos, creating chaptered summaries that significantly reduced the staff’s information consumption time.

Improved the efficiency of the Verbal Fluency Pipeline by integrating a speech-to-text pipeline and used knowledge graph embeddings to analyze semantic clustering in category fluency tasks, improving detection of cortical dysfunction Worked under Pedro Pinheiro-Chagas and Jet Vonk.

-Led advanced backend development of the company’s upcoming car rental platform using Java, integrating multiple PostgreSQL databases with techniques like data partitioning and optimized HTTP request handling through endpoint structuring and response caching strategies.-Developed a comprehensive security framework for a multi-service application, implementing OAuth 2.0, JWT authentication, and managing RESTful APIs with Spring Security, while successfully integrating multiple microservices and a gateway service.-Innovated in authentication and API integration with a multi-factor system, integrating third-party APIs for payment, and developed an e-billing system, ensuring efficient and secure asynchronous communication and transaction processing.

The Mango Jelly is under the UC Berkeley SkyDeck Accelerator and it helps users/businesses to experience the Collaborative AI Workspace and Copilot crafted specially for marketing teams. -Created a Django-based sales copilot leveraging natural language understanding techniques to translate user queries into API calls for various data platforms, achieving an 18% increase in data retrieval efficiency-Leveraged Lightwood Technology and Hugging Face Transformers, integrating MindsDB AI models with SQL to craft time series and classification models for sales forecasting, resulting in a substantial 108% increase in company sales.

-Developed a system utilizing credit/debit mechanisms using Python, facilitating end users in monitoring their bank transactions through the employment of ERC-20, ERC-1155, and NFT token transactions.-Worked on creating a rule-based identification model in order to analyze the inputs of transactions interacting with the smart contract and compute similarities between the input and other inputs in the dictionaries that store smart contracts-Utilized JSON, REST, and various APIs including PolygonScan API to compile data from the Ethereum database effectively.