Harnoor Dhingra
a human being who thinks, writes, talks, and codes. sometimes all at the same time.
My mom and dad are both doctors - say hi to ! Growing up, I'd have long discussions with them about why they were so wary of using tech in healthcare. Their answer was always some version of: it's hard to trust something you don't fully understand. Which, honestly? Fair. That's what put me on this path; I genuinely believe in building AI with a human lens.
I think a lot about multi-perspective generation, retrieval, bias, stereotype modeling, and personalization, especially how these interact inside large language systems. There's so much world knowledge encoded in these models, yet the hardest questions remain: how do you surface what you actually need? Whose ideas does a model flatten? Are we all eventually going to converge our thinking to one? If you think about these things too, reach out - always happy to talk!
Right now I'm at Microsoft in Cambridge, MA working on really cool problems with some of the best people.
I grew up in the northern state of Punjab in India, where community is everything - and that stayed with me. I write poetry about the life stuff, attend live events including concerts, and romanticize the quotidian.
- Deconstructing Identity Bias & Stereotypes in LLMs - oral at ACL Workshop 2023. How do you get a language model to see people with more regard? Chain-of-thought prompting + SHAP works.
- Multimodal Deep Learning for Personality Trait Prediction - Hierarchical co-attention over speech & text for finding personality in a voice.
Carnegie Mellon University - MS Artificial Intelligence, 2023. Made
BITS Pilani, Goa - BE Computer Science, 2021. Found my here.
