What's Next: Vision Engineering at Indra Group
My current role as a Senior Vision Engineer at Indra Group, building AI-powered vision systems for defense and aerospace, and how the research years led here.
Since December 2025 I’ve been a Senior Vision Engineer at Indra Group, back in Córdoba, designing and developing computer vision solutions for defense and aerospace applications. It’s the role that, in hindsight, the previous few years were quietly building toward.
Same disclaimer as always with defense work: I can talk about the shape of what I do, not the specifics of any one system. So consider this less “here’s the project” and more “here’s what the job actually is.”
From markers to mission-critical
A Ph.D. spent on camera pose estimation and fiducial markers turns out to translate surprisingly directly into “AI-powered vision systems for mission-critical environments”: the math doesn’t change just because the stakes did. What’s different is the environment: research code has to convince a reviewer and reproduce a result; this has to run reliably in the kind of setting where “it mostly works” isn’t a sentence anyone wants to hear.
Day to day, that means image processing and software engineering work sitting closer to systems engineering than to a research pipeline: less “can this be done at all,” more “can this be trusted every single time.”
Cameras, real products
A good chunk of the work is integrating industrial cameras, Keyence, IDS, Hikrobot, among others, into products that actually ship, rather than into a one-off research rig. On top of that hardware layer, I build vision checking systems: automated inspection products that catch what a manual check would miss. Some of that is for other clients’ projects, some of it is for defense, and same disclaimer as above applies to the latter: I naturally can’t say much about it.
Focus areas
Tech stack
The general-purpose computer vision toolkit, applied in a much less forgiving setting:
C++
Python
OpenCV
CMake
GitWhy I like it
Vision engineering in this space sits exactly at the intersection I’ve always liked best: real computer vision problems, applied to something that actually matters operationally, with none of the “solve it once for a paper and move on” pressure that academic research runs on. The bar for a system like this is that it keeps working, quietly, for years, which is a genuinely different and satisfying kind of hard.
This is very much a current chapter, not a finished one. I expect this post to look understated in a year or two once I can say more about what’s actually been built.