Products we've designed and built
A look at some of the AI, computer vision and web platforms we've shipped — each built from the ground up.
Every project on this page is an independent, in-house build — designed, engineered and shipped by CodeVar Technologies rather than delivered for a client under NDA. We are a young company and would rather show you work we can talk about openly, in full technical detail, than list logos we cannot explain.
Taken together they cover the range we work across: large language models and retrieval-grounded assistants, real-time computer vision on live camera feeds, voice interfaces, and conventional multi-user web platforms with payments, dashboards and reporting. The stacks span Python, Django, PyTorch and OpenCV on the AI side and Next.js, Node.js, PostgreSQL and AWS on the web side — which is deliberate. Choosing the right tool for a problem is only possible if you have actually built with several.
What each project solves
- Custom GPT
Repetitive knowledge work — the same domain questions and near-identical documents, every week. Built on OpenAI's GPT Builder and grounded in uploaded reference files, so there is no backend, vector database or hosting to maintain.
- SafeVision
Safety monitoring that depends on people watching camera feeds does not scale, so hazards get noticed only in the footage review after an incident. OpenCV and PyTorch models turn the cameras already installed into an always-on detection layer.
- Friday AI
Everyday automation scattered across a dozen apps, each needing a screen and a tap. Speech-to-text plus an LLM interpreting intent puts all of it behind one spoken interface, with an extensible skill system for new capabilities.
- WasteMantra
Waste collection still coordinated over phone calls and paper records, with no shared view of what was collected. A Next.js, Node.js and PostgreSQL platform on AWS gives customers and collectors one trackable source of truth.
How we approach a build
The pattern is the same across all four: start from the friction a real person experiences, pick the smallest architecture that removes it, and ship something working early enough to be corrected. SafeVision is containerised so it can run on edge hardware near the cameras instead of requiring a cloud round-trip. Custom GPT deliberately has no backend at all, because the problem did not need one. That restraint is a design decision, not a shortcut.
The same thinking applies to client work. If you have a problem worth solving, see the services we offer or tell us about it — the first conversation is about whether the thing should be built, not what it would cost.