🏡

I am Research Scientist at the Center for Science of Science and Innovation working for Dashun Wang at the Kellogg School of Management, Northwestern University. As a Research Scientist, I oversee the lab's overall AI efforts, data analytics, and data management. I work closely with other team members to ensure these initiatives align with the lab's goals and contribute to its broader research objectives. My research interests lie in the intersection of Large Language models, Science of Science, Representational Learning, and Uncertainty Quantification.

Previously, I was a Ph.D student advised by Dr. Hamed Alhoori, and my dissertation fully funded by the NSF grant. focussed on building a predictive modeling framework to investigate the Reproducibility crisis in AI. Previously, I was a Givens Research Associate and an Argonne Leadership Computing Facility Graduate Student researcher.

My most recent projects include training reasoning models and doing tests on scaling hypothesis for test time compute for scientific text generation. Specifically, finetuning and performance tuning DPO, ORPO, and recently GRPO on different open weight models. More can be found here. Additionally, I'm interested in conducting research investigating the topic of credit assignment problem for qualitative texts in science (peer-reviews). More specifically, if there exists a scientific task, Creating reward signals for LLM reasoning beyond math/programming domains is hard is a well agreed upon notion. More about this can be found in project LMRSD.

• 🔭 I’m currently working on fine-tuning large language models, implementing LLM agents, LLM's + Graphs.
• 🌱 I’m currently exploring Ways bring GraphRAG to the larger scholarly ecosystem, and Science of Science data
• 👯 I’m looking to collaborate on Building microservices for LLMs; Agentic LLM tooling as a service layer
• 🤔 I’m always looking for help with Building Knowledge Graphs within Citation Networks
• 💬 Ask me about Large Language models, Graph Learning, Uncertainty Quantification, Neural Architecture Search.
• 📫 How to reach me: @akhilpandey95
• 😄 Pronouns: (he/him)
• ⚡ Fun fact: Haskell has type inference, meaning it can automatically determine the type of a data by looking at how it is created.