Bio

šŸ‘‹ Hi there. This is Viju Sudhi (ą“µą“æą“œąµ ą“øąµą“§ą“æ in Malayalam).

I’m currently working as a Research Associate / PhD student at the Semantic Computing Group in Bielefeld University under the supervision of Prof. Dr. Philipp Cimiano.

šŸ§‘ā€šŸ’» A bit about myself

Developing a strong outlook in various areas of Natural Language Processing and Generation. Experienced in building and evaluating Generative Dialog Systems, fine-tuning Large Language Models and Retriever models.

✨ What am I currently working on?

Currently, I am working mostly on the project LLM4KMU - LLMs for Small and Medium-sized Enterprises.

LLM4KMU aims to adapt and optimize Large Language Models for their application in SMEs. As part of the project, we are building AutoLLM, an experimentation platform which supports users in finding the right open source LLM, fine-tunes the selected models, conducts extensive post-training and evaluation runs - finally supporting the user in making a more informed decision about the models they use for their application.

In addition to the development of the platform, we are also actively researching into the following topics:

Hallucination Detection / Mitigation – e.g. How can inference-time / training-free decoding strategies help mitigate hallucinations?

Resolving Knowledge conflicts in Language Models – e.g. Can LMs reliably over-ride parametric knowledge when presented with conflicting in-context knowledge?

Question Answering with Small Language Models – e.g. Can SLMs perform well when trained on domain-specific QA tasks?

āœšŸ» If you find common research interests, I am happy to collaborate! 😊

Please feel free to connect with me at viju.sudhi@uni-bielefeld.de.

šŸŽ§ What am I listening to (now)?