Discrete Conditional Generative Models for In-Silico Antibody Variant Design

Ongoing Project: Gflownets for phenomics guided small molecules discovery• Supervised by Michał Koziarski and Yoshua BengioPrevious Project: Greedier Gflownets using mixture policy with RL• Implemented various RL baselines such as DQN, SAC, A2C, and QGFN• Setup task environments (ex: bit sequences) and ran hyperparameter search experiments• Made improvements to cleanup codebase and facilitate visualizing and running experiments• Co-author on ICML paper: QGFN: Controllable Greediness with Action Values• Supervised by Elaine Lau and Emmanuel Bengio

Project: Evaluating the benefits and limitations of LLMs in clinical context• Setup a conversational Q&A agent with chat history and semantic document retrieval in LangChain.• Deployed a vector database using Zep with ingestion pipeline for adding new searchable text documents.• Built a dashboard & chat interface in SvelteKit for patients to interact with the augmented language model.Supervised by Dr. Devin Singh

• Designed a server-side rendering library that produces interactive React components from static JSON schemas, enabling the team to deliver near real-time frontend updates through a low-code platform.• Hired and onboarded three full-time frontend developers and organized weekly scrum, sprints, and code reviews.• Deployed dashboards, mobile app, and web app to 80 000+ yearly patient users and 5 000+ hospital staff users.

Project: Exploring the transcriptional mediators of stress in the nucleus acumbens• Clustered single-cell RNA-seq data in Scanpy using KNN, NMF, and Louvain methods.• Produced heatmap, violin, and dot plot visualizations to identify differentially expressed genes• Trained multiview pose estimation models on mice videos using DeepLabCut