A selected list of my projects in the open space.
A selected list of my projects in the open space.
Democratizing high-tech lab instrumentation with transformers
Analytical chemistry traditionally requires expensive tools and manual fragment annotations failing entirely on out-of-distribution data.
End-to-end encoder-decoder transformer that translates tandem mass spectra
Demonstrates that Test-time tuning is a dynamic adaptation strategy allowing the pre-trained model to update its weights at inference time
Making sense of data in the wild
33% of R&D and enterprise AI budgets are burned in data pipelines
Focus: Engineering automated, high-throughput data ingestion architectures
Agentic Orchestration appraoch involving LLM and AI agents that autonomously analyze raw, unstructured data across public repositories.
Learning from negative experimental results
Trial-and-error chemistry generates abundant "failed" experiments, but traditional datasets discard them due to success bias, wasting critical information about the boundaries of chemical space.
Used RL pipelines to teach LLMs to learn from negative data such as failed chemical reactions and High Throughput Experimentation data
Integrating negative chemical data drastically boosts the robustness and accuracy of reaction prediction models.