I am a Founding Research Engineer at Voyage AI, where I led the development of the first Instruction-Following rerankers (voyage-Rerank 2.5) in the Industry, achieveing state-of-the-art performance on numerous benchmarks. I now drive research on next-generation multimodal embedding models, pushing the frontier of cross-modal understanding and retrieval.

I graduated from Carnegie Mellon University (CMU) with an M.S. in Intelligent Information Systems (Dec 2024). I worked with Prof. Paul Pu Liang and Prof. LP Morency on improving cross-modal interactions in Large Vision-Language Models. As a part of my reseach, we released (HEMM: Holistic Evaluation of Multimodal Foundation Models) a holistic evaluation framework that systematically assesses multimodal foundation models across diverse tasks by decomposing performance based on varying degrees of multimodal interaction and reasoning complexity. My research interests broadly lie in Multimodal Understanding, Post-training, Reasoning and Continual Learning.

• Released the first Instruction Following re-ranker: voyage-rerank-2.5.
• Our work "HEMM: Holistic Evaluation of Multimodal Foundation Models" accepted for publication at NeurIPS 2024.

• Liang, Paul Pu*, Akshay Goindani*, Talha Chafekar, Leena Mathur, Haofei Yu, Ruslan Salakhutdinov, and Louis-Philippe Morency. "HEMM: Holistic Evaluation of Multimodal Foundation Models." Advances in Neural Information Processing Systems 37 (NeurIPS 2024): 42899-42940. [PDF]
• Sivaprasad Sarath*, Akshay Goindani*, Mario Fritz, and Vineet Gandhi. "Class-wise Domain Generalization: A Novel Framework for Evaluating Distributional Shift." NeurIPS 2022 Workshop on Distribution Shifts: Connecting Methods and Applications. [PDF]
• Akshay Goindani, Manish Shrivastava. "A Dynamic Head Importance Computation Mechanism for Neural Machine Translation." International Conference on Recent Advances in Natural Language Processing (RANLP 2021). [PDF]
• Kodali, P., Sachan, T., Goindani, A., Goel, A., Ahuja, N., Shrivastava, M. and Kumaraguru, P. "PreCogIIITH at HinglishEval: Leveraging Code-Mixing Metrics & Language Model Embeddings To Estimate Code-Mix Quality." arXiv preprint arXiv:2206.07988 (2022). [PDF]