Computer Science & Computer Vision AI/ML Intern

WHAT YOU OWN

The sensor hears a drone before anyone sees it. Your job is the part that turns what it hears, and what the cameras see, into a call the operator can act on. You own a real slice of the perception stack, with your own dates, alongside the engineers. You report to the CTO.

• Computer vision on the edge. Detection and tracking of Group 1-3 airframes in visible and infrared imagery. Training, evaluation, and the honest confusion matrix that comes with them. Small, fast, low-power models, not leaderboard models. The electro-optical and infrared line is a development track, not a shipped product, and you would be working on it from the beginning.
• Deployment on NVIDIA Jetson. Getting a model off your laptop and onto the unit: quantization, TensorRT, throughput and latency measured on the actual board under Caribbean heat, not estimated.
• Dataset and labeling pipeline. Collection, labeling protocol, versioning, augmentation, train and validation splits that do not leak. The training corpus is a company asset and you help build it.
• Field data capture. You come to the range. Live airframes, fixed and moving, multiple types. You help run the collection and you carry the gear.
• Evaluation you would defend. Precision, recall, false alarm rate, and the conditions each was measured under. Written so it can drop into a federal proposal without rewriting.
• Fusion with the RF side. Working with the RF engineer so that what the antenna detects and what the camera sees agree on a single track. Both halves are passive. The system listens and looks; it never transmits.
• Tooling. The small scripts, dashboards, and harnesses that make the above repeatable by someone who is not you.

The model you ship is the model that runs on the unit in the field. Nobody downstream second-guesses it. That is the weight of the seat.

HOW WE WORK

We are small, and we move at a pace that only a small team can. Engineering runs on two-week cycles against a single shared plan. Every seat, RF, mechanical, electrical, and software, owns a slice of the same rig and has to land its objectives on the same date the others do. An antenna that is finished three weeks before the enclosure is not finished. A mount that is finished after the field test is a failure.

You scope your own work, commit to dates, say early and loudly when a date is at risk, and integrate with the rest of the stack instead of handing over a part and walking away. Pace is a requirement of the seat, not a culture slogan, and it is the reason a company our size can put fielded hardware in front of a government customer.

WHAT WE REQUIRE

Pursuing or holding a B.S. in Computer Science, Computer Engineering, or a related field. This is an early-career seat and we expect to teach you things. What we cannot teach in six months is whether you build. Hands-on work counts wherever you did it: thesis, senior design, student competition, a personal project, or a bench in your apartment. We care that you have done it, not who paid you to.

• You have trained a model and looked hard at where it failed. Any domain. Tell us what the model got wrong, why, and what you changed. A project where everything worked tells us nothing.
• Python, working fluency. NumPy, and PyTorch or TensorFlow. You can read someone else's repository and get it running.
• Computer vision fundamentals. Convolutional networks, object detection and tracking, data augmentation, overfitting, the difference between validation and test. OpenCV.
• Linux comfort. Shell, ssh, git. You can work on a headless machine without a GUI.
• You finish things. One project, start to finish, that you can put on a screen and walk us through. Coursework counts. So does a bench in your apartment.
• Willing to be outside. Field days are part of this seat, not an exception to it.
• Juvempleo eligibility. Ages 18 to 29, Puerto Rico resident, U.S. citizen, minimum 2.00 GPA, recently graduated or about to graduate. Placement is made by DDEC, not by ITERRA alone.
• Method over conclusions. If the measurement says the product underperforms, we need to hear it from you first, and we will act on it, not argue with it.
• U.S. person under ITAR.
• Onsite in Hatillo, Puerto Rico, and in the field, outdoors, carrying gear.

Preferred: multi-object tracking. EO/IR or thermal imagery. Model optimization (TensorRT, ONNX, quantization, pruning). ROS. C++. Drone or UAS hands-on time. Clearance eligibility. Spanish.

WHAT WE OFFER

This seat is funded through Juvempleo, the Departamento de Desarrollo Economico y Comercio youth employment program. DDEC carries the wage. What ITERRA gives you is the work itself.

• $15.00 per hour, paid through Juvempleo. 950 hours, roughly six months.
• Real hardware from week one. You are not shadowing anybody. You get a slice of the rig, a date, and the same standard of review the engineers get.
• Direct access to the CTO and the founder. No layers. Your work is reviewed by the people who will ship it.
• Your work goes to the field. Whatever you build gets carried outside, mounted, and run against live targets while you are still here. Very few engineers get that before their first full-time job.
• A portfolio you can show anybody. Code, measurements, and a documented result you were responsible for, on a fielded defense system built in Puerto Rico.
• First look at a full-time seat. We hire from this bench. Performing here is the shortest path onto the payroll.