Launch HN: OctaPulse (YC W26) – Robotics and computer vision for fish farming
Source: Hacker News
Introduction
Hi HN! My name is Rohan and, together with Paul, I’m the co‑founder of OctaPulse (https://www.tryoctapulse.com/).
We’re building a robotics layer for seafood production, starting with automated fish inspection. We are currently deployed at our first production site with the largest trout producer in North America.
Why Aquaculture?
You might be wondering how we got into this with no background in aquaculture or the ocean industry.
- Coastal roots – I’m from Goa, India; Paul is from Malta and Puerto Rico.
- Cultural ties – Seafood is deeply tied to both our cultures and communities.
- Environmental concern – We saw firsthand the damage being done to our oceans and how wild fish stocks are being fished to near extinction.
- Global nutrition – Fish is the main protein source for almost 55 % of the world’s population.
- Import paradox – The U.S. imports 90 % of its seafood. That felt absurd and became the initial motivation for starting this company.
The Origin Story
Paul and I met at an entrepreneurship happy hour at CMU. In a land‑locked city like Pittsburgh there aren’t many people who care about the ocean, let alone want to build a company around it.
- We talked about ocean tech for three hours.
- I was drawn to building in the ocean because it is one of the hardest engineering domains out there.
- Paul had been researching aquaculture for months and kept finding the same thing: a $350 B global industry with less data visibility than a warehouse.
After that conversation we knew we wanted to work on this together.
The Problem: Manual Hatchery Workflows
Hatcheries – the early‑stage, on‑land part of production – are full of labor‑intensive workflows that are perfect candidates for automation.
- Farmers need to measure their stock for feeding, breeding, and harvest decisions, but fish are underwater and get stressed when handled.
- Most farms still sample manually: they net a few dozen fish, anesthetize them, place them on a table, measure one by one, and extrapolate to populations of hundreds of thousands.
- It takes ≈ 5 minutes per fish and the data is sparse.
When we saw this process we were baffled – there had to be a better way. This was the starting point that really kicked us off.
Core Challenges
- Harsh environment – Humidity, salt water, and corrosion are the enemy of anything mechanical.
- Underwater computer vision – Turbidity, particles, unpredictable fish motion, deformation, constant occlusion, and uncontrolled lighting make perception hard.
- Handling live fish – Fish are slippery, fragile, and stress easily; all materials must be food‑safe.
Vision System
- Cameras: Luxonis OAK cameras (depth + RGB) in a compact form factor.
- On‑board inference: Myriad X VPU runs lightweight models (detection, tracking) directly on the camera, avoiding constant USB frame streaming.
- Heavier workloads: Nvidia Jetson family (Orin Nano, Orin NX) for segmentation and key‑point extraction, chosen based on power and thermal constraints at each site.
Models & Optimization
| Task | Architecture | Deployment Runtime |
|---|---|---|
| Detection | YOLO variants | TensorRT / ONNX Runtime |
| Segmentation | Custom heads (CNN/Transformer) | TensorRT (INT8) |
| Key‑point extraction | Transformer‑based | OpenVINO / TensorRT |
- Quantization: INT8 on TensorRT gives the speed we need, but we must guard against accuracy loss, especially for segmentation boundaries.
- Calibration dataset: Captures lighting changes, water clarity shifts, and varying fish density; essential for robust quantized models.
Connectivity & Edge Computing
- Most farms have no Wi‑Fi; we use Starlink for connectivity in remote/offshore locations.
- All inference runs locally; data syncs only when a connection is available.
- We do not stream video to the cloud.
Internal Tooling
Early attempts with off‑the‑shelf labeling platforms fell short because they didn’t fit our workflow. We built a custom system that:
- Assigns labeling tasks to annotators.
- Tracks progress and versions datasets.
- Pushes models to edge devices with a single command.
This tight integration lets us close the loop quickly: data collection → labeling → training → quantization → deployment.
Robotics Hardware
- Enclosures: Custom‑built around off‑the‑shelf components to survive wet and humid conditions.
- Delta robots: Started with the Delta X S as a test platform. We are evaluating whether to move to industrial delta robots or design our own once we validate kinematics and payload requirements.
- End‑effector: Soft‑robotics compliant grippers (vacuum and typical grippers fail in this environment).
- Must handle fish of varying sizes, shapes, and life stages without damage.
- Design is still evolving; we are iterating on adaptive, food‑safe gripper mechanisms.
Current Focus
We are presently concentrating on operations outside the water – automating the sampling, measurement, and data capture steps in hatcheries.
If you’re interested in the technical details, collaborations, or just want to chat about ocean tech, feel free to reach out!
Overview
Chery phenotyping, sorting, quality inspection – these approaches are far more accessible than full‑underwater deployment and cheaper to start with. By combining genetics data, environmental data, and phenotypic imagery, we can help farms identify which fish to breed and which to cull. This is where selective breeding begins.
Why It Matters
- Genetic improvement is rare in aquaculture – only a tiny fraction of farmed fish species have gone through genetic improvement programs.
- Contrast with other livestock – chickens now grow 4× faster than they did in 1950 thanks to decades of selective breeding.
- Current state of fish farming – most farmed fish are essentially wild genetics.
The opportunity to improve aquaculture genetics is massive, but it is completely bottlenecked on measurement. You cannot improve what you cannot measure, and farms can barely measure anything at scale so far.
Trust Is the Gatekeeper
The industry moves on trust. We’re dealing with live animals, and farms are cautious about who they let near their stock. Coming from outside aquaculture, that trust had to be earned.
- Initial credibility – Paul was already a Future Leader with the Coalition for Sustainable Aquaculture.
- Breakthrough moment – attending the World Aquaculture Society (the largest conference in the US) led to a connection with the incoming lead geneticist at what became our first customer.
- Result – that relationship turned into a paid pilot with the largest trout producer in North America.
The Team
- Dan – background at ASML, Nvidia, Tesla, and Toyota.
- Paul – former Bloomberg employee.
- Both – met at CMU and immediately decided to tackle this problem and devote our lives to it.
Call for Feedback
We would love input from anyone who has experience with:
- Computer vision in harsh or unpredictable environments
- Edge deployment on constrained hardware (e.g., Jetson, OAK cameras)
- Gentle and appropriate handling of live animals with robotics
- Quantization workflows for inference
If you have aquaculture experience, we’re curious about problems we haven’t yet encountered.
Demo (NDAs Apply)
We can’t share demo videos due to NDAs, but here’s a photo of us building our initial dataset for phenotyping and morphometric analysis:
Join the Conversation
This is a weird industry to be building in, and we are learning something new every week. If you have experience with edge deployment, robotics in wet environments, or aquaculture itself, we would love to hear your perspective. And if you just have questions about fish or the tech, we’re happy to go deep in the comments. Excited to hear what this community thinks.
- Comments URL:
- Points: 9
- Comments: 0