Low-level engineering
Code that runs close to the hardware: C, Rust, inline assembly, direct work with memory and peripherals.
Needed where cycles and bytes matter more than framework convenience: drivers, firmware, speed-critical paths.
Technologies
What we actually work with
Not a list of buzzwords, but tools we run in production — from mounting a rack and configuring VLANs to fine-tuning models and writing code that sits close to the metal.
Low-level and systems programming
Rust
A systems language with no garbage collector: memory safety is verified at compile time.
Services that hold under load and don't leak. Our security policy controller is written in Rust.
Go
A compiled language with cheap concurrency and predictable resource usage.
High-load APIs and network services that are easier to operate.
Infrastructure, networks and operations
Bare metal and server hardware
Server sourcing and delivery, rack mounting, cabling, UPS, RAID, server operating systems.
Infrastructure doesn't start in the cloud — it starts with a physical machine someone has to install properly.
Networks and routing
Structured cabling, fibre, VLANs, VPN, seamless Wi-Fi. MikroTik, Cisco, Ubiquiti UniFi, pfSense.
A segmented network is both a performance decision and the first line of security.
Proxmox / VMware
Virtualisation on your own hardware: Proxmox VE, VMware ESXi, Hyper-V, backups and disaster recovery.
A dozen isolated environments on one server — and a recovery plan for when it fails anyway.
Kubernetes
Container orchestration: self-healing, autoscaling, zero-downtime rollouts.
The product survives a node failure and a traffic spike without an engineer on night duty.
ArgoCD
GitOps delivery: the cluster state lives in a repository and ArgoCD reconciles reality with it.
Every deploy is a commit. You see who changed what and when, and a rollback takes seconds.
Terraform
Infrastructure as code: servers, networks and access described declaratively and version-controlled.
An environment is reproduced from scratch in minutes instead of being assembled from memory.
Monitoring and alerting
Zabbix, Prometheus, Grafana: hardware and service metrics, logs, dashboards, incident alerts.
An engineer learns about the problem before the client does — and before it becomes downtime.
Self-hosted solutions
Corporate services on your own infrastructure: mail, file storage, Git, CI, VPN.
Data stays inside the perimeter and the bill doesn't grow with every new employee.
Docker
Packaging applications and dependencies into a reproducible image.
"Works on my machine" stops being an argument: it behaves the same everywhere.
Nginx
Reverse proxy, load balancing, TLS termination, static delivery.
An entry point that holds thousands of connections on modest hardware.
Data and storage
PostgreSQL
A relational database with transactions, JSONB, extensions and honest data consistency.
The primary store in almost all of our projects: money and stock levels must not be lost.
ClickHouse
A columnar analytical database: aggregates over billions of rows in seconds.
Dashboards and reporting that don't take down the production database with heavy queries.
Redis
In-memory store: cache, queues, sessions, counters, locks.
Takes repetitive load off the database and makes the interface feel instant.
MinIO / S3
Object storage with an S3-compatible API — on your own infrastructure or in the cloud.
Files, backups and media live apart from the app and don't hit the server's disk limit.
MongoDB
A document database for data with a shifting structure.
Fits where the schema hasn't settled yet or differs from record to record.
Artificial intelligence
On-premise LLM
Deploying open models (Llama, Qwen and others) on the client's own infrastructure.
The language model works with internal documents and not a single request leaves the perimeter.
RAG
Retrieval-Augmented Generation: the model answers from your knowledge base rather than from memory.
An assistant over company documents that cites its source and invents far less.
Fine-tuning
Additional training of models on the data and terminology of a specific domain.
The model learns your documents, catalogue and wording instead of the generic internet.
NLP
Natural language processing: classification, entity extraction, document parsing.
Incoming mail, requests and contracts are parsed automatically, without manual sorting.
Computer vision
Object detection and classification, image and video stream analysis.
Quality control on the floor, document recognition, analytics from cameras.
Speech technologies
Speech recognition and synthesis, transcription and audio processing.
Calls and meetings turn into text you can search and analyse.
Security
Zero Trust
No implicit trust: a two-phase handshake (X25519 ECDH + Ed25519) and a three-tier JWT system.
Breaching one service doesn't open the whole system. This is how our BMSA product is built.
HashiCorp Vault
Central secret storage: keys, passwords and tokens with rotation and access audit.
Secrets don't sit in environment variables and don't leak together with the repository.
WireGuard
A modern VPN built on elliptic-curve cryptography: fast, compact, easy to audit.
Remote access to the company perimeter without heavy legacy solutions.
mTLS and OAuth 2.0
Mutual certificate authentication between services, OAuth 2.0 / JWT for users, RBAC and MFA.
Every internal call is verified — and it's clear exactly who made it.
Product engineering
Python
Django and FastAPI on the backend, plus the entire ML toolchain.
One language covers both business logic and work with data and models.
Next.js
React, TypeScript, server rendering, Tailwind CSS.
Interfaces that load fast and are indexed correctly by search engines.
React Native
iOS and Android from a single codebase, with native modules where needed.
Two app stores without two separate development teams.
Telegram Bot API
Bots, Mini Apps and notifications wired into CRM, ERP and payment systems.
The client talks to the business where their messenger is already open.
We choose the stack for the task, not the other way round. Below are the tools we run in production.
Next step
Tell us about your project
Describe the task in the form — we will come back with clarifying questions, a scope outline and a rough estimate. Even if you only have an idea and no specification yet.
The brief takes about 3 minutes to fill in.