01
The work
I led a ten-person undergraduate research team through iterative protein-expression work spanning literature search, plasmid and process heuristics, parameter selection, protocol execution, and measured output.
The project received CURC Highest Honors for its system-level experimental design and workflow. The public companion preserves the project context and interactive research surface.
02
AI was a throughput layer, not the verifier
We used AI to widen parameter search, organize literature, and make protocol decisions easier to inspect. That let a young team explore more of the design space without pretending the generated recommendation was the result.
The bench remained authoritative. Expression had to survive the protocol, the instrument, the controls, and the team’s interpretation of physical measurements.
03
What carried forward
This project gave me the pattern I now see everywhere: computation proposes, a carefully designed process acts, an external instrument measures, and the next decision inherits the trace.
Agentic scientific discovery becomes credible when it preserves that separation while making the loop faster.