Is it even worth it
I help figure out two things: where AI actually pays off, and where you're better off not starting. If the idea is sound, I help get it to something that runs. If it isn't, I say so up front — not six months in.
Down
4+
years messing with AI, edge and robots
3
pieces of hardware taken from idea to production
100s
camera + compute units running out in cities
on-device
neural nets running on the device itself, no server
What I actually do
Most companies start with AI from the wrong end. Short version: I help start from the right one and get it to something that works. Here's what that means in practice.
How much hardware, and what it costs
I work out what it'll run on: your own servers, GPUs, cloud. Honest numbers — what to buy and what it costs to keep running. If you can't put a number on the cost of inference, the project isn't ready yet.
Where AI will actually help
I look at how things are set up and find the spots where an LLM, computer vision or agents earn their keep. Usually the problem isn't the model, it's the data — so that's where I start. I rarely suggest agents as a first step.
Local or cloud
If the data can't leave the building, we do it locally. If local gets too expensive, we price the cloud. If it's expensive everywhere, I tell you that before we start, not after.
Whether it'll pay for itself
I work out roughly what it'll save or earn before the money is spent. Sometimes the answer is that it's not worth it — that's a fine outcome too, and better to know early.
Neural nets on the device itself
When data can't leave the device, or the cloud bill gets silly, I run the models locally: on-device, edge, your own LLM. I did exactly this on drones and cameras, where there simply is no server.
The model is maybe ten percent of the work. The rest is data, hardware, and keeping it from falling apart in the field.
Aleksandr Davydov
Engineer · computer vision, edge AI, robots
Who I am
Engineer. Four-plus years messing with full-stack, embedded, edge AI and robots — and getting all of it to actually work out in the field.
I started as a developer who took on whatever a project needed. Then I became the one who runs the team and owns what actually ships: secure government platforms, camera-detection products on city streets, AI for autonomous drones, rovers and submarines.
In Abu Dhabi I ran the R&D group for autonomous robots. Before that, at City Technologies, we built three products that went from idea to hardware — they're on the streets now across Russia and neighbouring countries.
Over the years I learned one simple thing: the model is maybe ten percent of the work. The rest is data, hardware, integration, and keeping it from falling apart a month later in the field. That's what I talk about with people trying to fit AI into their business.
- Where I am
- Saint Petersburg · work from anywhere
- Languages
- Russian (native) · English (working)
From the work
Honestly, this is more interesting than a headshot. Boards, cameras, Jetsons and drones — the stuff I actually put together by hand and got working.
Cameras on the roads
Roadside camera + compute units that recognise vehicles, plates and events in real time. Running in several cities.
Machine-vision cameras
Picked and assembled the multi-camera capture for the onboard rig — detection and tracking on the move.
A robot with compute on board
A Jetson on a robot doing detection and scene understanding fully offline — no server anywhere.
Drones
Ground control and vision for autonomous drones. The hard part is spotting a small, low-contrast target from hundreds of metres up.
That board in the cat case
When there's no off-the-shelf hardware, you make your own. I laid out and brought up this board myself — the cat enclosure I printed just for fun.
The stuff that's actually on the street
Several hundred units shipped and kept running in the real world, not in the lab.
Where I worked
Where all this came from.
On my own now
Jun 2026 — nowHelping companies with AI · Remote / worldwide
- I help figure out where AI pays off and get sensible ideas to something that runs.
- I work out the hardware and the cost of inference; local, edge or cloud — whichever comes out cheaper and safer.
- I say so honestly when a project is better left alone.
Black Diamond Research & Development
Dec 2024 — Jun 2026Head of R&D group · Abu Dhabi, UAE
- Ran development, requirements and the team for R&D on autonomous drones, rovers and submarines.
- Built the onboard vision: several cameras, detection and tracking in real time — including the hard long-range, low-contrast targets.
- Got multimodal models running straight on Jetson Orin and RK3588, no server.
City Technologies LTD
May 2022 — Dec 2024Tech lead, computer-vision team · Belgorod / Moscow
- Ran the team; three products went from idea to hardware that's on the streets now across Russia and neighbouring countries.
- Camera detection, mobile detection on the move, city parking — vehicles, people and events in real time.
- Shipped and kept running several hundred camera + compute units.
Belgorod Regional Government
Aug 2021 — Feb 2022Lead developer (contract) · Belgorod
- Built a secure internal communications platform for government officials.
- The whole thing: requirements, security, encrypted document workflow, internal messaging.
What I work with
The tools, if you're curious.
AI / ML
- LLM & multimodal
- RAG
- Fine-tuning & LoRA
- Quantization
- Computer vision
- Detection / YOLO
- Tracking & re-ID
- Agents
Edge & embedded
- NVIDIA Jetson Orin
- Jetson Thor
- Rockchip RK3588
- TensorRT
- DeepStream
- on-device inference
- Embedded Linux
Everything else
- C / C++
- Python
- Rust
- TypeScript / React
- PostgreSQL
- Docker & CI/CD
- Linux
- MLOps
Write to me if it's relevant
In a couple of lines — what you're trying to do. I'll tell you honestly whether AI is even the right tool, what it would take, and roughly what it costs. If you don't need AI, I'll say that too.
PDF · in English and Russian