Rajdeep Mondal
Examine this. Do not believe it.
I am Rajdeep Mondal, Senior Data Scientist and GenAI Lead at Elucidata. I build AI systems for biology, and the job reduces to one demand: the machine defends its answers to scientists. Scientists check claims for a living. Building for them ruined easier audiences for me.
The path here was not a plan. I traded markets before I built models, and trading taught me position size and how to be wrong fast. I read code, autobiographies, and systems books in whatever order curiosity picked them. Each thread looked like a distraction until the work needed all of them at once. The longer project is founder judgment: problem selection, distribution, writing, proof.
Outside the work I walk long approaches in the Himalaya. Four treks so far, and the descent is still the goal. In person I am messier than this page, and quicker to say when I am not sure.
Every claim below has a public link, a number I stand behind, or a note that the work is private.
Three public results.
- 01I won first place in the NCI AI Data Readiness Challenge, $20,000, at 95% accuracy.
- 02I released 23,650 cancer patient phenopackets with NCI and Lawrence Berkeley Lab. Oncopacket on GitHub (opens in a new tab)
- 03I am first author on a multi-agent curation system that reaches 93% recall across 23 metadata fields. Preprint on bioRxiv (opens in a new tab)
How I work.
Work clears three gates before it gets my time. The wrong answer has to cost something real, because interesting is not enough. The system has to show its sources, and it has to stop when its confidence is dishonest. And the problem comes before the model, because technical depth is the entry point rather than the destination.
With people, the call comes first and the argument comes second. When the argument fails, the call changes in the open. Claims arrive with receipts: a number, a mechanism, or a test. When I do not know, I say so and name the test that finds out.
The pace depends on the door. Reversible decisions get decided fast and corrected cheap. Irreversible ones get slowed down and priced first.
I am blunt about the work and never about the person. Expect to hear where the plan fails, with the failure priced. Evidence moves me faster than pressure does, and a wrong claim gets corrected in the same channel, at the same volume.
Decisions go on paper before the work starts: the call, the odds, the condition that reverses it. The record is how the next bet improves. Hold me to every line of this.
How to write to me.
The address goes to me, and nobody screens it. Expect a reply in the same voice you just read.
