AIZOYA · Applied AI research

Research

Exploring how AI can help people find evidence, understand complex data, and solve unfamiliar problems.

Completed research projects

Biohub · Kaggle · 2026

Tracking life with AI

Founder-reported completed entry · Research

Our competition research explored detecting cells, linking their movement, and identifying divisions in 3D time-lapse microscopy of zebrafish development.

Illustrative research overview. No prize has been reported by the founder. It is not a diagnostic tool or a clinically validated product.

Read the video transcript

Life moves. One cell at a time. But following those cells through three-dimensional microscopy is a complex challenge.

Which cell is which? Where did it move? And how does its path connect across time?

We explored that challenge through Kaggle’s Biohub Cell Tracking During Development competition.

Our public competition score improved from a 0.827 baseline to a verified 0.953 submission.

Behind those numbers are experiments, difficult decisions, and lessons that shape what we build next.

This project shows how we approach research: define the challenge, test carefully, measure progress, and learn from the evidence.

Explore the project and what we learned.

The 0.953 public score came from a reproduction of a public competition notebook, not a claim of a novel model, accuracy percentage, final rank, or award.

Video not playing? Open the video ↗

Explore the official challenge ↗

Kaggle · KaggricultureSubmitted entry · Founder-reported

FarmBrain · Planning under limited resources

Our competition work explored AI strategies for crop planning, resource allocation, and decisions in a simulated farming environment. This participation does not claim an award or verified agricultural performance outside the simulation.

NIH / National Library of Medicine · SPARK

Biomedical discovery

Franklyn reports submitting registration requests for four SPARK biomedical research challenge tasks. These are planned research directions; we will share solutions and results if and when they are ready.

dbGaP · Task 1Request submitted · Founder-reported

Making research data easier to compare

Our planned focus: matching inconsistent study-variable names to standard biomedical terminology.

dbGaP · Task 2Request submitted · Founder-reported

Finding relevant studies in plain language

Our planned focus: helping researchers discover suitable studies through conversational search of study metadata.

PubMed/PMC · Task 1Request submitted · Founder-reported

Finding the right scientific source

Our planned focus: retrieving the publication that answers a specific biomedical question, with a traceable source.

PubMed/PMC · Task 2Request submitted · Founder-reported

Exploring evidence with citations

Our planned focus: organizing answers to broader research questions around supporting scientific publications.

Solution/code deadline: January 15, 2027. See the official task pages for submission stages and updates. Status reviewed September 28, 2026.

Challenge registration does not establish NIH funding, partnership, endorsement, or clinical validation.

Research in progress

DrivenData · GEMS PrizeResearch preparation

Finding hidden geological faults

We’re exploring machine-learning approaches to geological fault identification using the GEMS challenge data. The challenge aims to improve mapping that supports geothermal-resource exploration.

Early research preparation using the GEMS challenge data. This page reports exploratory work, not a validated model or submitted solution.

AIcrowd · Alignment Research CenterResearch in progress

ARC White-Box Estimation Challenge 2026

Can we estimate a neural network’s average internal activations from its weights more efficiently than repeatedly running it? We’re investigating mathematical estimators under a fixed compute budget.

Our work focuses on reproducible baselines, numerical verification, and independent technical review. Results are still under evaluation.

Competition research

Reasoning in unfamiliar environments

Investigating how AI agents explore, learn, and solve interactive tasks they have not encountered before through ARC-AGI-3.

Follow the work

Explore our demonstrations, competition entries, and earned platform milestones.

View competition activity