Year One at Loopbio

A year ago I wrote, in a post called Another Change , that I was leaving the comfort of academia (mostly) to go and build the tools I kept wishing existed. I closed by observing that I was apparently an entrepreneur now, a sentence I had not expected to write, and asked you to wish us luck. You did, and here is the honest accounting of what a year of that luck looked like.

It looked, mostly, like building. The pitch deck version of co-founding Loopbio is all vision and slides with the word “platform” on them; the lived version is a great deal more soldering. Most of the year went into the unglamorous middle of it: camera-synchronisation boards, tested in arrays of ten because one is never enough; sturdy frames in black anodised T-slot extrusion, because everything we build, we build to last (short-term solutions have a way of staying with you a very long time, and long-term ones a way of never happening); and lighting rigs, which sound trivial right up until you discover that nearly every tracking and image-processing problem is, underneath, a lighting problem wearing a disguise. The arrival of a Digikey box, I can confirm, is genuinely like Christmas for engineers.

Which is where the joke on me lives. In that post I wrote, with some conviction, that the apparatus should not be the hard part - the science should be the hard part. I still believe every word of it. I had simply not appreciated that the way you arrange for the apparatus not to be the scientist’s hard part is by making it our hard part, in perpetuity, with a soldering iron. The thesis, it turns out, was a job description.

We did get out of the workshop occasionally. I spent a few days at the VISION fair in Stuttgart among the 10GigE and NBASE-T cameras (5G and 2.5G now, over more than ten metres of cable, which solves a real problem), lenses the size of small dogs, and more 3D and time-of-flight imaging - more shiny tech per square metre - than is entirely good for the soul. I kept half an eye on the research frontier too, and read the NIPS write-ups with interest. And yes, a Titan X Pascal turned up and was, for about a week, the most exciting object in the office, which tells you something about the office. Deep learning is a real and useful tool, and we are investing in the compute to do it properly; I remain, as ever, wary of treating a useful descriptor as a magic one.

The part that actually justifies the year, though, is the scientists. Simon Gingins has been sending us collective-behaviour footage of fish from the Red Sea, off Eilat, in such volume that field work there evidently comes with excellent internet; we have been building wand autocalibration so that he can have that behaviour in three dimensions rather than the flattering lie of a single camera. There has been fish VR with the collective-behaviour people, zebrafish tracking with real-time object detection, and 3D cup tracking creeping forward calibration sphere by calibration sphere. They are getting on with the science while we sweat the apparatus, which is precisely, exactly, the point of the company. The first time a scientist sends you their data instead of their bug reports, you understand what the company was for.

It has been harder than the slides implied - more customer support, more soldering, more standing in fluorescent hall after fluorescent hall - but it is unmistakably the right kind of hard, the kind where the difficulty is in the work and not in wondering whether the work matters. So: thank you to the collaborators who trusted a one-year-old company with their data, and to Max, with whom none of this would work. It was a good 2016, and we go into 2017 still meaning to make your research dreams a reality, robot arm or no robot arm.