Shipping inspectionPowders, semiconductor materials
Check whether the particle size of these 3 lots is within spec
Verdict
L-103 is out of spec
L-101, L-102All particles 40µm or lessL-1035 of 75 particles exceed 40µm
Example (illustrative)
GeXeL Agent
From images to
verdicts, causes,
and improvements.
Bubbles, cells, particles.
Measured one by one, answered with evidence.
What is GeXeL Agent?A next-generation platform for manufacturing that lets you use GeXeL from your own generative AI
Before / After
After you take the image, the work changes
Before
Until now
Every step by hand
After
With GeXeL Agent
People just take images and ask
1Take images
2Measure with software
3Export to a spreadsheet and tally
4Make charts
5Compare with past datato find differences
6Plan the next conditions
1Take images,then ask generative AI
Check how this image differs from the previous one
GeXeL Agent takes over
Measure each one
Tally the results
Make charts
Show differencesin numbers
Suggest next conditions
Use cases
Judgments differ by person
01＼ Quality Control ／
What should we try next…
Can’t reproduce it at scale
03＼ Manufacturing ／
01Quality Control
Make everyday quality
visible
Complaint handlingBatteries
Compare the lot reported for “fast capacity loss” with a normal lot
Comparison result
4% → 17%
Share of cathode particles with cracks.Normal lot3 of 75Complaint lot13 of 75
Example (illustrative)
02R&D
Both performance and patents,
from numbers in images.
Performance improvement (informatics)Healthcare
Suggest the next culture conditions from how cell shape relates to growth
Culture conditions to try next
Oxygen 5% · 3,000 cells/cm²
Oxygen concentration and cell seeding density.Conditions with more elongated cells showed better growth.Predicted growth1.3× the best so far
Example (illustrative)
Patent claimsChemicals
Show how our foam differs from 3 competitor products, in numbers we can use in a patent claim
Proposed numerical range for the patent claim
Average bubble size 50µm or less, and circularity 0.85 or more
Combining the two conditions, only our product falls within the range.
Example (illustrative)
03Manufacturing
Images show
why anomalies occur
Scale-upCosmetics, food
Compare oil droplet sizes in the lab and in mass production, and suggest production conditions
Homogenization pressure and passes to try next
70MPa · 3 passes
Average oil droplet size measured from imagesLab (target)1.2µm9 production trials1.6–3.4µmPredicted for this condition1.3µm
Example (illustrative)
Causes of defectsPharmaceuticals
Compare the cracked lots with good ones, together with the process records
Possible cause
Low granule moisture
Granule moisture before tableting3 lots with many cracks1.1–1.3%Other 9 lots1.9–2.5%Only the lots with many cracks are below 1.5%.
Example (illustrative)
Across departments
Images from other departments
become evidence, too.
Quality Control 
Manufacturing data01Investigate the cause of a defective lot in Quality Control using Manufacturing data
Manufacturing R&D test conditions02Find scale-up conditions for Manufacturing from R&D test conditions
R&D QC customer feedback03Set numerical targets for R&D from customer needs heard by Quality Control
GeXeL
Measurements come from GeXeL image analysis
100+companies
GeXeL users to date
800+models
Pre-trained for analysis
As of the end of September 2026
Frequently asked questions
QHow is it different from the current GeXeL?
A
The analysis itself is the same GeXeL. The difference is that you can make requests in a conversation with generative AI, and receive the results as tables or reports.
QCan I use it without expertise in AI or image analysis?
A
Yes, you can. Just make requests in everyday language, such as “Check whether the particle size of these 3 lots is within spec.”
Unlike analysis software, there is no need to set conditions and run the analysis yourself. GeXeL Agent analyzes the images with an analysis model suited to them and answers based on the measured values.
QWhat should I do when the analysis results don’t match what I see?
A
Tell it in the conversation what looks different. There is no need to remove particles one by one by hand: criteria for excluding false detections (such as size or shape) are set, the matching detections are removed in one go, and the count is redone.
The exclusion criteria are kept on record, so you can check later which criteria the judgment was based on.
QCan we try it with our own images?
A
Yes, you can try it with images you already have. Please contact us via “Try with your own images” at the bottom of this page. We also welcome inquiries about early adoption.
Note that the images analyzed during use are uploaded directly to GeXeL, without passing through the generative AI service.
Start with your own images
You can try it with images you already have.