AI in FinanceTime Series

Forecasting Asset Connectedness

Working on AI in Finance where we forecast connectedness between any two assets, aimed at supporting better policy decisions in future unexpected or crisis events.

NLPLLM Fine-tuning

Romanized Telugu Language Model

Fine-tuned GPT-2 and evaluated mGPT for next-word prediction on Romanized, code-mixed Telugu WhatsApp chat data. Fine-tuning cut perplexity 4x, but the multilingual mGPT still beat it on accuracy (15% vs 4% top-1) due to GPT-2's tokenizer fragmenting Telugu words.

MLResearch

Project Title Three

Brief description of what this project does, the problem it solves, and any interesting technical details.