Instafill.ai joined NVIDIA Inception, NVIDIA's program for AI startups
NVIDIA accepted Instafill.ai into NVIDIA Inception on September 29, 2026, two weeks after we applied. Inception is NVIDIA's free program for startups building with AI: members get technical resources, training, preferred pricing on NVIDIA hardware and software, and introductions to investors. NVIDIA reviews each application before accepting a company. For us the fit is practical, because one of the two AI steps in Instafill.ai already runs on an NVIDIA GPU.
At a glance
- Program NVIDIA Inception, NVIDIA's free program for AI startups
- Applied September 15, 2026
- Accepted September 29, 2026
- Where NVIDIA runs in Instafill.ai The field detector behind flat-to-fillable conversion, on an NVIDIA GPU since October 12, 2026
- Detection time About 0.3 seconds a page, down from 11–26 seconds
Where Instafill.ai already runs on NVIDIA
Most forms people bring us are flat: a scan or a PDF with no fields at all. Before anything can be filled, flat-to-fillable conversion has to find where the fields are. A computer vision model looks at each page and marks every text box, checkbox and signature area; language models then name the fields and decide what each one asks for.
Since October 12, 2026, that detection model runs on an NVIDIA GPU instead of the CPU server it was hosted on before. Measured on the same forms before and after the switch:
| CPU server, before | NVIDIA GPU, since October 12 | |
|---|---|---|
| Field detection, per page | 11–26 seconds | about 0.3 seconds |
| A three-page conversion, start to finish | 62–98 seconds | 23 seconds |
Tested on the same pages, the GPU returns the same fields as the CPU server did. The CPU server stays on as a backup, so a page is still converted if the GPU cannot answer. Nothing changed for anyone using Instafill.ai except the wait: you can see it on the free fillable PDF tool, which runs the same conversion.
What the membership gives us
What NVIDIA offers Inception members, and what we intend to use it for:
| Benefit | What we use it for |
|---|---|
| Preferred pricing and rebates on NVIDIA hardware and software, plus cloud credits from NVIDIA Cloud Partners | More GPU capacity for field detection as conversion volume grows |
| Free self-paced courses and discounted instructor-led workshops from NVIDIA Training | Skills for training and serving computer vision models |
| The NVIDIA Developer Program: SDKs, APIs, developer tools and documentation | Running our models faster on NVIDIA GPUs |
| AI model APIs optimized for NVIDIA hardware | Trying new models before we build anything on them |
| Introductions to NVIDIA's partners and venture capital firms | Partnerships and funding as the company grows |
What we are working on next
Speed was the first win. The next ones are about accuracy, and they come to the web app, the Agent, the API and MCP at the same time:
- Better field detection on scanned and low-quality forms, where faded lines, stamps and skewed pages make fields harder to see.
- A more precise AI field review, the step that removes extra boxes a detector draws on a page.
Our mission is the same as on day one: nobody should have to retype information into forms by hand, and any form, even a scan with no fields, can be turned into a fillable form and filled with AI from your own documents.
