Co-founder & CTO, Computer Vision (United States)

About Homaya

Homaya Inc. is an early-stage technology company building retail technology with computer vision. See homaya.ai.

We are launching in the US with quick-service restaurants, convenience stores and grocery. We are looking for a technical co-founder to own the technology, end to end, from the first store to the hundredth.

The role

You own everything technical, and you build it with your own hands first.

• Vision in production. Detection, multi-object tracking and re-identification on real store cameras, at many sites.
• Vision-language models. Fine-tune, distil and calibrate compact VLMs so their outputs can be trusted in production.
• Edge. Inference on a small device in the store: PyTorch to a C++ runtime, TensorRT or ONNX Runtime, FP16 and INT8, many camera streams per box, privacy masking on the device.
• Fleet. Provision, update over the air and monitor devices across hundreds of stores.
• Cloud and data. Video and event pipelines on AWS, built to pass a SOC 2 review.
• Precision. Own the false-positive rate. Every alert a store sees has to be worth their time.
• Data engine. Evaluation sets, shadow mode and a loop that learns from real stores with few labels.
• Team. Hire and lead the engineering team after funding, with a high bar.
• Customers. Sit with operators in pilot stores and turn what you see into the product.

Must have

All of these, with something shipped behind each one.

1. You have shipped computer vision to production on CCTV or fixed cameras, live across multiple sites.
2. You have built detection, tracking and re-identification yourself, not only managed people who did.
3. You have run VLMs in production: fine-tuning (LoRA or similar), distillation and calibrated outputs. You know when a VLM should verify and when a detector should decide.
4. You have taken PyTorch models to the edge: C++ runtime, TensorRT or ONNX Runtime, FP16 or INT8, profiled for latency and throughput on Jetson, OpenVINO or Hailo, with many RTSP streams per device.
5. You own the end metric, not the model. You can name the false-positive rate of a system you shipped, and explain how you cut it.
6. You have built something from zero before: a startup, or a new product inside a company.
7. You ship fast with AI-assisted development and know where it breaks.
8. You can commit full-time, and you want to be a founder, not an employee.

Nice to have

• Vision work in retail, restaurants, loss prevention or workplace safety
• Integration with NVRs and VMS platforms (ONVIF, RTSP)
• Device fleets: provisioning, over-the-air updates, Yocto, remote monitoring
• AWS, Kubernetes and infrastructure as code (Terraform or Pulumi)
• SOC 2, CCPA or BIPA, and on-device privacy by design
• Go or C++ for backend and real-time systems
• Turning research papers into working code
• Time inside a strong engineering organisation of 50 or more engineers
• Synthetic data or rare-event training
• Technical selling alongside a founder: demos, security reviews, pilot readouts

Not a fit

• Research only, with nothing running in production
• Cloud-only inference, never on a device
• Results only on public benchmarks or demo videos
• Looking for a part-time, fractional or advisory role
• Looking for a salaried job now

The deal

• Founder equity , a meaningful co-founder stake with standard four-year vesting and a one-year cliff.
• Salary from funding. Until then, this is an equity role.
• A short trial first. We work together on a real problem for a few weeks before equity is issued, so both sides know.
• Location: US preferred. Remote works if you overlap US hours.

How to apply

No CV. Send a short note to View email address on apply.placed-app.com or a LinkedIn DM. Start the subject line (or the first line of your DM) with HMY-CTO-SHIPPED, then your name. Notes without it are not read. Include:

1. The last vision system you shipped: cameras, sites, hardware and its false-positive rate
2. One link to code, a paper or a product you built
3. Where you are based and when you could start

We reply to every note that answers all three.