Founder · Researcher · Builder

Rishabh Goel

I build things that make life a little smarter.

Building Bond Health Based Boston, MA Prev. Doorstep.ai · Regeneron

Scroll to begin

Nº 00 Prologue

Hey, I'm Rishabh. A founder and applied-ML researcher, and a compulsive builder of things that, with any luck, make life a little smarter.

Right now I'm building Bond Health, an AI company I co-founded and run as CEO. We use large language models to read electronic health records and match patients to the clinical trials they actually qualify for, turning a months-long, multi-billion-dollar bottleneck into something that takes weeks. Alongside it, I'm finishing a Master's in Biomedical Informatics at Harvard Medical School.

Before Bond, I co-founded Doorstep.ai (last-mile delivery, computer vision), taught LLMs to read single-cell genomics at Regeneron, and built clinical ML at GEn1E Lifesciences that earned a $749K federal grant. Along the way I've published in NEJM AI, filed patents, and was named one of Boston's 25 Under 25.

But the part I actually care about is simpler: hard problems, fast feedback loops, and building things people genuinely use.

Rishabh GoelFIG. 01

THE WORKING THEORY

Hard problems. Fast feedback loops.
Things people actually use.

Machine LearningLarge Language ModelsClinical NLPKnowledge GraphsComputer VisionNeural ODEsSingle-cell RNA-seqBioinformaticsPyTorchTensorFlow Machine LearningLarge Language ModelsClinical NLPKnowledge GraphsComputer VisionNeural ODEsSingle-cell RNA-seqBioinformaticsPyTorchTensorFlow

Nº 01 The Work

Things I've built.

Five chapters at the intersection of AI, product, and biomedicine, from clinical language models to last-mile logistics.

Co-founder & CEO

2025 - Present

LLMsClinical NLPKnowledge GraphsPython

Bond Health ↗

Getting patients into the right trials.

The company I co-founded and run as CEO. Bond is an LLM-native platform that reads structured EHR signals and free-text clinical notes to match patients to the trials they actually qualify for, then carries them through to informed consent. It's grounded in a clinical knowledge graph of 1.6M+ nodes and backed by Plug and Play, Leo Capital, and CEAS Investments.

90%+trial-match accuracy vs. a ~67% public baseline

Machine Learning Intern

Summer 2025

LLMsSingle-cell RNA-seqPython

Regeneron ↗

LLMs for single-cell biology.

On Regeneron's Molecular Profiling & Data Science team, I built a Python package that wires proprietary LLMs into the company's single-cell RNA-seq workflow, turning a prompt and a dataset into automated biological summaries. It cut new-user onboarding by 80%+, reached full adoption across a 30+ researcher team, and beat baseline methods by 17% on clinical-accuracy benchmarks.

+17%over baseline · adopted across 3 departments

Co-founder & CIO

2023 - 2024

Computer VisionCNNsPyTorch

Doorstep.ai ↗

Finding the front door.

I co-founded Doorstep as CIO and built its early computer-vision stack: CNNs that pinpoint a delivery location to within half a meter, where ordinary GPS is roughly 8× worse, especially indoors. I moved on in 2024; the company has since raised an $8M seed (led by Canaan Partners) and now operates across the U.S.

0.5 mlocation accuracy · ~8× better than GPS

Rishabh pitching at the MIT $100K competition — On stage, the founder years.MIT $100K PITCH

Biomedical Data Scientist

2022 - 2024

Scikit-learnSVMBioinformaticsPython

GEn1E × BARDA

Sorting the unsortable.

At GEn1E Lifesciences (Y Combinator-backed), I built a support-vector-machine model that identifies clinical subtypes of ARDS from high-dimensional biomarker data, surfacing patient subgroups that respond differently to treatment. The work won a $749K federal grant from BARDA's DRIVe program, and the findings are under peer review.

97%+subphenotyping accuracy · $749K BARDA grant

Research Fellow

2021 - 2022

TensorFlowNeural ODEsNumPyPython

Merck × UC Berkeley

Predicting how a drug moves.

A Neural ODE model for predicting pharmacokinetics in patients: how a drug is absorbed, distributed, and cleared over time. It beat state-of-the-art methods by 4% and earned "Best Undergraduate Research Project" at ADSA 2023.

+4%over state-of-the-art · Best Undergrad Research, ADSA 2023

Nº 02 By the Numbers

$749K

BARDA research
grant

Patents
filed

4.0

GPA, Harvard
& Berkeley