Alibaba Cloud AI Catalyst Program · Application
An AI research analyst that turns medical researchers’ survey data into analysis, tables and a written paper.
Izaan Shaikh, Founder · kyrostat.com
Clinicians run surveys and studies, but turning the data into a correct analysis and a written Results section takes weeks, and usually a statistician they don’t have.
Choosing the right test, grouping answers, handling missing data.
Publication-ready, in the journal’s format.
Methods and Results that match the numbers exactly.
1 · Upload
The researcher uploads survey or study data.
2 · Ask
In plain language: “compare junior and senior surgeons on this outcome”.
3 · Kyro does it
Runs the tests, builds tables and figures, writes Methods and Results into the paper.
4 · Researcher decides
Every assumption is listed to confirm or correct; every change can be undone.
Live at kyrostat.com and in use by researchers preparing papers for journal submission.
Kyro plans each request as a checklist, writes Python, and runs it in a sandbox on the researcher’s own data.
Before finishing, it reviews every change against the request and states its assumptions for the researcher.
qwen3.7-plus runs the agent; a flash model handles quick suggestions. Prompt caching keeps long sessions affordable.
qwen3.8-flash and qwen3.8-max, measured on replays of real researchers’ requests.
230M
input tokens on qwen3.7-plus, most read from the prompt cache
3,900
model calls by the research agent and other AI features
[__]
researchers using KyroStat
Much of today’s spend is testing: we replay real researchers’ requests against the agent to measure and raise its accuracy before each release.
Source: Alibaba Cloud Model Studio usage and billing, 1–24 October 2026
Credits to fund testing and our first wave of researchers.
Guidance on the best Qwen model and pricing for an always-on research agent.
Co-marketing as a medical-research case study on Model Studio.
Izaan Shaikh, Founder · kyrostat.com · [your business email]