Niroshan Ranapathi is a Senior Full Stack Engineer at Temus, bringing over 5 years of experience in software development. He holds a Master's degree in Business Administration from the University of Wolverhampton and a Bachelor's degree in Computer Science and Engineering from the University of Moratuwa. Niroshan has previously worked as a Technical Specialist at Axcer and as a Senior Application Consultant at NCS Group. His expertise encompasses Full Stack Development, e-commerce, mobile application development, and database management, with proficiency in technologies such as AWS, Angular, and React. Currently, he is focused on developing innovative web and mobile applications, leveraging cloud-based architectures to implement complex systems.

Full Stack Development, Software Engineering, E-Commerce Development, Database Management, Mobile Application Development, AWS, Angular, React, Cloud Technologies

Currently developing full stack web and mobile applications at Temus using technologies like AWS, Angular, and React. Focusing on e-commerce and enterprise software solutions, implementing complex systems with cloud-based architectures. Actively contributing to software development projects that leverage modern web technologies and cloud infrastructure.

I am an AI engineer with 4 years of experience in machine learning and software development. I hold a masters degree in computer science from National university of Singapore. I have mainly worked in research labs and start ups and consistently tinker with the latest research developments in the field of AI. I have been working in the field since the GPT-3 (Ada cabbage Davinci days) and have experience in building applications on top of foundational models and scaling it to thousands of users. Additionally, I have lot of experience iterating on the model level where I have worked on finetuning open weights models. Some of the features I have built have directly been recognized and used by the US government.

AI Agents , RL for LLMs. , Synthetic Data, Privacy

I am currently working on finetuning multimodal llms for masking confidential data in documents. With the proliferation of vision language models, documents can be fed as images (page by page) and a decoder llm can be conditioned for extracting confidential data from each page. This setup ensures that data in charts/images etc also get captured as opposed to converting documents into text and then feeding it to LLMs. (This approach always has context loss)