Built an OCR pipeline in Python for handwritten medical records using Llama, Gemini, GPT, and Qwen vision models. The work supports record digitization in Sri Lanka, with field deployment planned in Uganda.
Yash Sharma / Personal site
I like moving
people towards
ideas.
Research. Music. Ideas.
Keep moving ↓IDEAS → PEOPLE → MOMENTUM → IDEAS → PEOPLE → MOMENTUM →
01 / Work
Reached more than 500 programmers. Built over 20 curricula and ran hackathons, workshops, and 24-hour Accelerator challenges. Put in over 2,000 hours of work.
Access site ↗Co-led a six-member officer team, planned speaker events, and ran a donation drive that delivered supplies to more than 150 people. Managed recurring meetings averaging 15 attendees and grew the club’s social media presence.
02 / Research
Annals of Emergency Medicine / Vol. 84, Issue 4EMERGENT: Emergency Medicine Resources and Guidelines Enhanced by Natural Language Text-Generation
Y. Sharma · D. Dash — Built a RAG system that answers emergency medicine questions using accepted clinical policies and published evidence, and shows its sources.
Read abstract ↗Deep learning / PyTorchCancer Classification with T-GEMBuilt and trained a T-GEM transformer in PyTorch to classify cancer subtypes from miRNA data.
View project ↗Machine learning / June 2025Foundational Modeling of Molecular Self-Assembly DynamicsBuilt a GNN + LSTM model for molecular self-assembly. It reached 0.00448 MSE after 238 epochs, ran 51× faster than molecular dynamics simulations, and improved on prior baselines by 2×.
View project ↗03 / Listen
Podcast / SpotifyMachine Learning
Artist / SpotifyTHESPRITZ04 / Contact